efbf5322d70377f218b82a02089e5973c08cfec4
[libav.git] / libavcodec / h264.c
1 /*
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 *
19 */
20
21 /**
22 * @file h264.c
23 * H.264 / AVC / MPEG4 part10 codec.
24 * @author Michael Niedermayer <michaelni@gmx.at>
25 */
26
27 #include "common.h"
28 #include "dsputil.h"
29 #include "avcodec.h"
30 #include "mpegvideo.h"
31 #include "h264data.h"
32 #include "golomb.h"
33
34 #include "cabac.h"
35
36 #undef NDEBUG
37 #include <assert.h>
38
39 #define interlaced_dct interlaced_dct_is_a_bad_name
40 #define mb_intra mb_intra_isnt_initalized_see_mb_type
41
42 #define LUMA_DC_BLOCK_INDEX 25
43 #define CHROMA_DC_BLOCK_INDEX 26
44
45 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46 #define COEFF_TOKEN_VLC_BITS 8
47 #define TOTAL_ZEROS_VLC_BITS 9
48 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49 #define RUN_VLC_BITS 3
50 #define RUN7_VLC_BITS 6
51
52 #define MAX_SPS_COUNT 32
53 #define MAX_PPS_COUNT 256
54
55 #define MAX_MMCO_COUNT 66
56
57 /**
58 * Sequence parameter set
59 */
60 typedef struct SPS{
61
62 int profile_idc;
63 int level_idc;
64 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
65 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
66 int poc_type; ///< pic_order_cnt_type
67 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
68 int delta_pic_order_always_zero_flag;
69 int offset_for_non_ref_pic;
70 int offset_for_top_to_bottom_field;
71 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
72 int ref_frame_count; ///< num_ref_frames
73 int gaps_in_frame_num_allowed_flag;
74 int mb_width; ///< frame_width_in_mbs_minus1 + 1
75 int mb_height; ///< frame_height_in_mbs_minus1 + 1
76 int frame_mbs_only_flag;
77 int mb_aff; ///<mb_adaptive_frame_field_flag
78 int direct_8x8_inference_flag;
79 int crop; ///< frame_cropping_flag
80 int crop_left; ///< frame_cropping_rect_left_offset
81 int crop_right; ///< frame_cropping_rect_right_offset
82 int crop_top; ///< frame_cropping_rect_top_offset
83 int crop_bottom; ///< frame_cropping_rect_bottom_offset
84 int vui_parameters_present_flag;
85 AVRational sar;
86 int timing_info_present_flag;
87 uint32_t num_units_in_tick;
88 uint32_t time_scale;
89 int fixed_frame_rate_flag;
90 short offset_for_ref_frame[256]; //FIXME dyn aloc?
91 int bitstream_restriction_flag;
92 int num_reorder_frames;
93 }SPS;
94
95 /**
96 * Picture parameter set
97 */
98 typedef struct PPS{
99 int sps_id;
100 int cabac; ///< entropy_coding_mode_flag
101 int pic_order_present; ///< pic_order_present_flag
102 int slice_group_count; ///< num_slice_groups_minus1 + 1
103 int mb_slice_group_map_type;
104 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
105 int weighted_pred; ///< weighted_pred_flag
106 int weighted_bipred_idc;
107 int init_qp; ///< pic_init_qp_minus26 + 26
108 int init_qs; ///< pic_init_qs_minus26 + 26
109 int chroma_qp_index_offset;
110 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
111 int constrained_intra_pred; ///< constrained_intra_pred_flag
112 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
113 int transform_8x8_mode; ///< transform_8x8_mode_flag
114 }PPS;
115
116 /**
117 * Memory management control operation opcode.
118 */
119 typedef enum MMCOOpcode{
120 MMCO_END=0,
121 MMCO_SHORT2UNUSED,
122 MMCO_LONG2UNUSED,
123 MMCO_SHORT2LONG,
124 MMCO_SET_MAX_LONG,
125 MMCO_RESET,
126 MMCO_LONG,
127 } MMCOOpcode;
128
129 /**
130 * Memory management control operation.
131 */
132 typedef struct MMCO{
133 MMCOOpcode opcode;
134 int short_frame_num;
135 int long_index;
136 } MMCO;
137
138 /**
139 * H264Context
140 */
141 typedef struct H264Context{
142 MpegEncContext s;
143 int nal_ref_idc;
144 int nal_unit_type;
145 #define NAL_SLICE 1
146 #define NAL_DPA 2
147 #define NAL_DPB 3
148 #define NAL_DPC 4
149 #define NAL_IDR_SLICE 5
150 #define NAL_SEI 6
151 #define NAL_SPS 7
152 #define NAL_PPS 8
153 #define NAL_PICTURE_DELIMITER 9
154 #define NAL_FILTER_DATA 10
155 uint8_t *rbsp_buffer;
156 int rbsp_buffer_size;
157
158 /**
159 * Used to parse AVC variant of h264
160 */
161 int is_avc; ///< this flag is != 0 if codec is avc1
162 int got_avcC; ///< flag used to parse avcC data only once
163 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
164
165 int chroma_qp; //QPc
166
167 int prev_mb_skipped; //FIXME remove (IMHO not used)
168
169 //prediction stuff
170 int chroma_pred_mode;
171 int intra16x16_pred_mode;
172
173 int top_mb_xy;
174 int left_mb_xy[2];
175
176 int8_t intra4x4_pred_mode_cache[5*8];
177 int8_t (*intra4x4_pred_mode)[8];
178 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
179 void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);
180 void (*pred8x8 [4+3])(uint8_t *src, int stride);
181 void (*pred16x16[4+3])(uint8_t *src, int stride);
182 unsigned int topleft_samples_available;
183 unsigned int top_samples_available;
184 unsigned int topright_samples_available;
185 unsigned int left_samples_available;
186 uint8_t (*top_borders[2])[16+2*8];
187 uint8_t left_border[2*(17+2*9)];
188
189 /**
190 * non zero coeff count cache.
191 * is 64 if not available.
192 */
193 uint8_t non_zero_count_cache[6*8] __align8;
194 uint8_t (*non_zero_count)[16];
195
196 /**
197 * Motion vector cache.
198 */
199 int16_t mv_cache[2][5*8][2] __align8;
200 int8_t ref_cache[2][5*8] __align8;
201 #define LIST_NOT_USED -1 //FIXME rename?
202 #define PART_NOT_AVAILABLE -2
203
204 /**
205 * is 1 if the specific list MV&references are set to 0,0,-2.
206 */
207 int mv_cache_clean[2];
208
209 /**
210 * number of neighbors (top and/or left) that used 8x8 dct
211 */
212 int neighbor_transform_size;
213
214 /**
215 * block_offset[ 0..23] for frame macroblocks
216 * block_offset[24..47] for field macroblocks
217 */
218 int block_offset[2*(16+8)];
219
220 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
221 uint32_t *mb2b8_xy;
222 int b_stride; //FIXME use s->b4_stride
223 int b8_stride;
224
225 int halfpel_flag;
226 int thirdpel_flag;
227
228 int unknown_svq3_flag;
229 int next_slice_index;
230
231 SPS sps_buffer[MAX_SPS_COUNT];
232 SPS sps; ///< current sps
233
234 PPS pps_buffer[MAX_PPS_COUNT];
235 /**
236 * current pps
237 */
238 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
239
240 uint16_t (*dequant4_coeff)[16]; // FIXME quant matrices should be per SPS or PPS
241 uint16_t (*dequant8_coeff)[64];
242
243 int slice_num;
244 uint8_t *slice_table_base;
245 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
246 int slice_type;
247 int slice_type_fixed;
248
249 //interlacing specific flags
250 int mb_aff_frame;
251 int mb_field_decoding_flag;
252
253 int sub_mb_type[4];
254
255 //POC stuff
256 int poc_lsb;
257 int poc_msb;
258 int delta_poc_bottom;
259 int delta_poc[2];
260 int frame_num;
261 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
262 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
263 int frame_num_offset; ///< for POC type 2
264 int prev_frame_num_offset; ///< for POC type 2
265 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
266
267 /**
268 * frame_num for frames or 2*frame_num for field pics.
269 */
270 int curr_pic_num;
271
272 /**
273 * max_frame_num or 2*max_frame_num for field pics.
274 */
275 int max_pic_num;
276
277 //Weighted pred stuff
278 int use_weight;
279 int use_weight_chroma;
280 int luma_log2_weight_denom;
281 int chroma_log2_weight_denom;
282 int luma_weight[2][16];
283 int luma_offset[2][16];
284 int chroma_weight[2][16][2];
285 int chroma_offset[2][16][2];
286 int implicit_weight[16][16];
287
288 //deblock
289 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
290 int slice_alpha_c0_offset;
291 int slice_beta_offset;
292
293 int redundant_pic_count;
294
295 int direct_spatial_mv_pred;
296 int dist_scale_factor[16];
297 int map_col_to_list0[2][16];
298
299 /**
300 * num_ref_idx_l0/1_active_minus1 + 1
301 */
302 int ref_count[2];// FIXME split for AFF
303 Picture *short_ref[32];
304 Picture *long_ref[32];
305 Picture default_ref_list[2][32];
306 Picture ref_list[2][32]; //FIXME size?
307 Picture field_ref_list[2][32]; //FIXME size?
308 Picture *delayed_pic[16]; //FIXME size?
309 Picture *delayed_output_pic;
310
311 /**
312 * memory management control operations buffer.
313 */
314 MMCO mmco[MAX_MMCO_COUNT];
315 int mmco_index;
316
317 int long_ref_count; ///< number of actual long term references
318 int short_ref_count; ///< number of actual short term references
319
320 //data partitioning
321 GetBitContext intra_gb;
322 GetBitContext inter_gb;
323 GetBitContext *intra_gb_ptr;
324 GetBitContext *inter_gb_ptr;
325
326 DCTELEM mb[16*24] __align8;
327
328 /**
329 * Cabac
330 */
331 CABACContext cabac;
332 uint8_t cabac_state[460];
333 int cabac_init_idc;
334
335 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
336 uint16_t *cbp_table;
337 int top_cbp;
338 int left_cbp;
339 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
340 uint8_t *chroma_pred_mode_table;
341 int last_qscale_diff;
342 int16_t (*mvd_table[2])[2];
343 int16_t mvd_cache[2][5*8][2] __align8;
344 uint8_t *direct_table;
345 uint8_t direct_cache[5*8];
346
347 uint8_t zigzag_scan[16];
348 uint8_t field_scan[16];
349 const uint8_t *zigzag_scan_q0;
350 const uint8_t *field_scan_q0;
351
352 int x264_build;
353 }H264Context;
354
355 static VLC coeff_token_vlc[4];
356 static VLC chroma_dc_coeff_token_vlc;
357
358 static VLC total_zeros_vlc[15];
359 static VLC chroma_dc_total_zeros_vlc[3];
360
361 static VLC run_vlc[6];
362 static VLC run7_vlc;
363
364 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
365 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
366 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);
367
368 static inline uint32_t pack16to32(int a, int b){
369 #ifdef WORDS_BIGENDIAN
370 return (b&0xFFFF) + (a<<16);
371 #else
372 return (a&0xFFFF) + (b<<16);
373 #endif
374 }
375
376 /**
377 * fill a rectangle.
378 * @param h height of the rectangle, should be a constant
379 * @param w width of the rectangle, should be a constant
380 * @param size the size of val (1 or 4), should be a constant
381 */
382 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
383 uint8_t *p= (uint8_t*)vp;
384 assert(size==1 || size==4);
385
386 w *= size;
387 stride *= size;
388
389 assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
390 assert((stride&(w-1))==0);
391 //FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
392 if(w==2 && h==2){
393 *(uint16_t*)(p + 0)=
394 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
395 }else if(w==2 && h==4){
396 *(uint16_t*)(p + 0*stride)=
397 *(uint16_t*)(p + 1*stride)=
398 *(uint16_t*)(p + 2*stride)=
399 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
400 }else if(w==4 && h==1){
401 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
402 }else if(w==4 && h==2){
403 *(uint32_t*)(p + 0*stride)=
404 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
405 }else if(w==4 && h==4){
406 *(uint32_t*)(p + 0*stride)=
407 *(uint32_t*)(p + 1*stride)=
408 *(uint32_t*)(p + 2*stride)=
409 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
410 }else if(w==8 && h==1){
411 *(uint32_t*)(p + 0)=
412 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
413 }else if(w==8 && h==2){
414 *(uint32_t*)(p + 0 + 0*stride)=
415 *(uint32_t*)(p + 4 + 0*stride)=
416 *(uint32_t*)(p + 0 + 1*stride)=
417 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
418 }else if(w==8 && h==4){
419 *(uint64_t*)(p + 0*stride)=
420 *(uint64_t*)(p + 1*stride)=
421 *(uint64_t*)(p + 2*stride)=
422 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
423 }else if(w==16 && h==2){
424 *(uint64_t*)(p + 0+0*stride)=
425 *(uint64_t*)(p + 8+0*stride)=
426 *(uint64_t*)(p + 0+1*stride)=
427 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
428 }else if(w==16 && h==4){
429 *(uint64_t*)(p + 0+0*stride)=
430 *(uint64_t*)(p + 8+0*stride)=
431 *(uint64_t*)(p + 0+1*stride)=
432 *(uint64_t*)(p + 8+1*stride)=
433 *(uint64_t*)(p + 0+2*stride)=
434 *(uint64_t*)(p + 8+2*stride)=
435 *(uint64_t*)(p + 0+3*stride)=
436 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
437 }else
438 assert(0);
439 }
440
441 static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
442 MpegEncContext * const s = &h->s;
443 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
444 int topleft_xy, top_xy, topright_xy, left_xy[2];
445 int topleft_type, top_type, topright_type, left_type[2];
446 int left_block[8];
447 int i;
448
449 //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
450 // the actual condition is whether we're on the edge of a slice,
451 // and even then the intra and nnz parts are unnecessary.
452 if(for_deblock && h->slice_num == 1)
453 return;
454
455 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
456
457 top_xy = mb_xy - s->mb_stride;
458 topleft_xy = top_xy - 1;
459 topright_xy= top_xy + 1;
460 left_xy[1] = left_xy[0] = mb_xy-1;
461 left_block[0]= 0;
462 left_block[1]= 1;
463 left_block[2]= 2;
464 left_block[3]= 3;
465 left_block[4]= 7;
466 left_block[5]= 10;
467 left_block[6]= 8;
468 left_block[7]= 11;
469 if(h->mb_aff_frame){
470 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
471 const int top_pair_xy = pair_xy - s->mb_stride;
472 const int topleft_pair_xy = top_pair_xy - 1;
473 const int topright_pair_xy = top_pair_xy + 1;
474 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
475 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
476 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
477 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
478 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
479 const int bottom = (s->mb_y & 1);
480 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);
481 if (bottom
482 ? !curr_mb_frame_flag // bottom macroblock
483 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
484 ) {
485 top_xy -= s->mb_stride;
486 }
487 if (bottom
488 ? !curr_mb_frame_flag // bottom macroblock
489 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
490 ) {
491 topleft_xy -= s->mb_stride;
492 }
493 if (bottom
494 ? !curr_mb_frame_flag // bottom macroblock
495 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
496 ) {
497 topright_xy -= s->mb_stride;
498 }
499 if (left_mb_frame_flag != curr_mb_frame_flag) {
500 left_xy[1] = left_xy[0] = pair_xy - 1;
501 if (curr_mb_frame_flag) {
502 if (bottom) {
503 left_block[0]= 2;
504 left_block[1]= 2;
505 left_block[2]= 3;
506 left_block[3]= 3;
507 left_block[4]= 8;
508 left_block[5]= 11;
509 left_block[6]= 8;
510 left_block[7]= 11;
511 } else {
512 left_block[0]= 0;
513 left_block[1]= 0;
514 left_block[2]= 1;
515 left_block[3]= 1;
516 left_block[4]= 7;
517 left_block[5]= 10;
518 left_block[6]= 7;
519 left_block[7]= 10;
520 }
521 } else {
522 left_xy[1] += s->mb_stride;
523 //left_block[0]= 0;
524 left_block[1]= 2;
525 left_block[2]= 0;
526 left_block[3]= 2;
527 //left_block[4]= 7;
528 left_block[5]= 10;
529 left_block[6]= 7;
530 left_block[7]= 10;
531 }
532 }
533 }
534
535 h->top_mb_xy = top_xy;
536 h->left_mb_xy[0] = left_xy[0];
537 h->left_mb_xy[1] = left_xy[1];
538 if(for_deblock){
539 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
540 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
541 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
542 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
543 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
544 }else{
545 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
546 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
547 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
548 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
549 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
550 }
551
552 if(IS_INTRA(mb_type)){
553 h->topleft_samples_available=
554 h->top_samples_available=
555 h->left_samples_available= 0xFFFF;
556 h->topright_samples_available= 0xEEEA;
557
558 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
559 h->topleft_samples_available= 0xB3FF;
560 h->top_samples_available= 0x33FF;
561 h->topright_samples_available= 0x26EA;
562 }
563 for(i=0; i<2; i++){
564 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
565 h->topleft_samples_available&= 0xDF5F;
566 h->left_samples_available&= 0x5F5F;
567 }
568 }
569
570 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
571 h->topleft_samples_available&= 0x7FFF;
572
573 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
574 h->topright_samples_available&= 0xFBFF;
575
576 if(IS_INTRA4x4(mb_type)){
577 if(IS_INTRA4x4(top_type)){
578 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
579 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
580 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
581 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
582 }else{
583 int pred;
584 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
585 pred= -1;
586 else{
587 pred= 2;
588 }
589 h->intra4x4_pred_mode_cache[4+8*0]=
590 h->intra4x4_pred_mode_cache[5+8*0]=
591 h->intra4x4_pred_mode_cache[6+8*0]=
592 h->intra4x4_pred_mode_cache[7+8*0]= pred;
593 }
594 for(i=0; i<2; i++){
595 if(IS_INTRA4x4(left_type[i])){
596 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
597 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
598 }else{
599 int pred;
600 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
601 pred= -1;
602 else{
603 pred= 2;
604 }
605 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
606 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
607 }
608 }
609 }
610 }
611
612
613 /*
614 0 . T T. T T T T
615 1 L . .L . . . .
616 2 L . .L . . . .
617 3 . T TL . . . .
618 4 L . .L . . . .
619 5 L . .. . . . .
620 */
621 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
622 if(top_type){
623 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
624 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
625 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
626 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
627
628 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
629 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
630
631 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
632 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
633
634 }else{
635 h->non_zero_count_cache[4+8*0]=
636 h->non_zero_count_cache[5+8*0]=
637 h->non_zero_count_cache[6+8*0]=
638 h->non_zero_count_cache[7+8*0]=
639
640 h->non_zero_count_cache[1+8*0]=
641 h->non_zero_count_cache[2+8*0]=
642
643 h->non_zero_count_cache[1+8*3]=
644 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
645
646 }
647
648 for (i=0; i<2; i++) {
649 if(left_type[i]){
650 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
651 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
652 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
653 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
654 }else{
655 h->non_zero_count_cache[3+8*1 + 2*8*i]=
656 h->non_zero_count_cache[3+8*2 + 2*8*i]=
657 h->non_zero_count_cache[0+8*1 + 8*i]=
658 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
659 }
660 }
661
662 if( h->pps.cabac ) {
663 // top_cbp
664 if(top_type) {
665 h->top_cbp = h->cbp_table[top_xy];
666 } else if(IS_INTRA(mb_type)) {
667 h->top_cbp = 0x1C0;
668 } else {
669 h->top_cbp = 0;
670 }
671 // left_cbp
672 if (left_type[0]) {
673 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
674 } else if(IS_INTRA(mb_type)) {
675 h->left_cbp = 0x1C0;
676 } else {
677 h->left_cbp = 0;
678 }
679 if (left_type[0]) {
680 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
681 }
682 if (left_type[1]) {
683 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
684 }
685 }
686
687 #if 1
688 //FIXME direct mb can skip much of this
689 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
690 int list;
691 for(list=0; list<1+(h->slice_type==B_TYPE); list++){
692 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
693 /*if(!h->mv_cache_clean[list]){
694 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
695 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
696 h->mv_cache_clean[list]= 1;
697 }*/
698 continue;
699 }
700 h->mv_cache_clean[list]= 0;
701
702 if(IS_INTER(top_type)){
703 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
704 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
705 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
706 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
707 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
708 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
709 h->ref_cache[list][scan8[0] + 0 - 1*8]=
710 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
711 h->ref_cache[list][scan8[0] + 2 - 1*8]=
712 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
713 }else{
714 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
715 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
716 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
717 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
718 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
719 }
720
721 //FIXME unify cleanup or sth
722 if(IS_INTER(left_type[0])){
723 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
724 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
725 *(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]];
726 *(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]];
727 h->ref_cache[list][scan8[0] - 1 + 0*8]=
728 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
729 }else{
730 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
731 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
732 h->ref_cache[list][scan8[0] - 1 + 0*8]=
733 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
734 }
735
736 if(IS_INTER(left_type[1])){
737 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
738 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
739 *(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]];
740 *(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]];
741 h->ref_cache[list][scan8[0] - 1 + 2*8]=
742 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
743 }else{
744 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
745 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
746 h->ref_cache[list][scan8[0] - 1 + 2*8]=
747 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
748 assert((!left_type[0]) == (!left_type[1]));
749 }
750
751 if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
752 continue;
753
754 if(IS_INTER(topleft_type)){
755 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
756 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
757 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
758 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
759 }else{
760 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
761 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
762 }
763
764 if(IS_INTER(topright_type)){
765 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
766 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
767 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
768 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
769 }else{
770 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
771 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
772 }
773
774
775 h->ref_cache[list][scan8[5 ]+1] =
776 h->ref_cache[list][scan8[7 ]+1] =
777 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
778 h->ref_cache[list][scan8[4 ]] =
779 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
780 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
781 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
782 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
783 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
784 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
785
786 if( h->pps.cabac ) {
787 /* XXX beurk, Load mvd */
788 if(IS_INTER(topleft_type)){
789 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
790 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
791 }else{
792 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
793 }
794
795 if(IS_INTER(top_type)){
796 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
797 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
798 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
799 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
800 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
801 }else{
802 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
803 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
804 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
805 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
806 }
807 if(IS_INTER(left_type[0])){
808 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
809 *(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]];
810 *(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]];
811 }else{
812 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
813 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
814 }
815 if(IS_INTER(left_type[1])){
816 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
817 *(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]];
818 *(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]];
819 }else{
820 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
821 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
822 }
823 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
824 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
825 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
826 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
827 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
828
829 if(h->slice_type == B_TYPE){
830 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
831
832 if(IS_DIRECT(top_type)){
833 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
834 }else if(IS_8X8(top_type)){
835 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
836 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
837 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
838 }else{
839 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
840 }
841
842 //FIXME interlacing
843 if(IS_DIRECT(left_type[0])){
844 h->direct_cache[scan8[0] - 1 + 0*8]=
845 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
846 }else if(IS_8X8(left_type[0])){
847 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
848 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
849 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
850 }else{
851 h->direct_cache[scan8[0] - 1 + 0*8]=
852 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
853 }
854 }
855 }
856 }
857 }
858 #endif
859
860 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
861 }
862
863 static inline void write_back_intra_pred_mode(H264Context *h){
864 MpegEncContext * const s = &h->s;
865 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
866
867 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
868 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
869 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
870 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
871 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
872 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
873 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
874 }
875
876 /**
877 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
878 */
879 static inline int check_intra4x4_pred_mode(H264Context *h){
880 MpegEncContext * const s = &h->s;
881 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
882 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
883 int i;
884
885 if(!(h->top_samples_available&0x8000)){
886 for(i=0; i<4; i++){
887 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
888 if(status<0){
889 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
890 return -1;
891 } else if(status){
892 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
893 }
894 }
895 }
896
897 if(!(h->left_samples_available&0x8000)){
898 for(i=0; i<4; i++){
899 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
900 if(status<0){
901 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
902 return -1;
903 } else if(status){
904 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
905 }
906 }
907 }
908
909 return 0;
910 } //FIXME cleanup like next
911
912 /**
913 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
914 */
915 static inline int check_intra_pred_mode(H264Context *h, int mode){
916 MpegEncContext * const s = &h->s;
917 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
918 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
919
920 if(mode < 0 || mode > 6) {
921 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);
922 return -1;
923 }
924
925 if(!(h->top_samples_available&0x8000)){
926 mode= top[ mode ];
927 if(mode<0){
928 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);
929 return -1;
930 }
931 }
932
933 if(!(h->left_samples_available&0x8000)){
934 mode= left[ mode ];
935 if(mode<0){
936 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);
937 return -1;
938 }
939 }
940
941 return mode;
942 }
943
944 /**
945 * gets the predicted intra4x4 prediction mode.
946 */
947 static inline int pred_intra_mode(H264Context *h, int n){
948 const int index8= scan8[n];
949 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
950 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
951 const int min= FFMIN(left, top);
952
953 tprintf("mode:%d %d min:%d\n", left ,top, min);
954
955 if(min<0) return DC_PRED;
956 else return min;
957 }
958
959 static inline void write_back_non_zero_count(H264Context *h){
960 MpegEncContext * const s = &h->s;
961 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
962
963 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
964 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
965 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
966 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
967 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
968 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
969 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
970
971 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
972 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
973 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
974
975 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
976 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
977 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
978 }
979
980 /**
981 * gets the predicted number of non zero coefficients.
982 * @param n block index
983 */
984 static inline int pred_non_zero_count(H264Context *h, int n){
985 const int index8= scan8[n];
986 const int left= h->non_zero_count_cache[index8 - 1];
987 const int top = h->non_zero_count_cache[index8 - 8];
988 int i= left + top;
989
990 if(i<64) i= (i+1)>>1;
991
992 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
993
994 return i&31;
995 }
996
997 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
998 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
999
1000 if(topright_ref != PART_NOT_AVAILABLE){
1001 *C= h->mv_cache[list][ i - 8 + part_width ];
1002 return topright_ref;
1003 }else{
1004 tprintf("topright MV not available\n");
1005
1006 *C= h->mv_cache[list][ i - 8 - 1 ];
1007 return h->ref_cache[list][ i - 8 - 1 ];
1008 }
1009 }
1010
1011 /**
1012 * gets the predicted MV.
1013 * @param n the block index
1014 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1015 * @param mx the x component of the predicted motion vector
1016 * @param my the y component of the predicted motion vector
1017 */
1018 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1019 const int index8= scan8[n];
1020 const int top_ref= h->ref_cache[list][ index8 - 8 ];
1021 const int left_ref= h->ref_cache[list][ index8 - 1 ];
1022 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1023 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1024 const int16_t * C;
1025 int diagonal_ref, match_count;
1026
1027 assert(part_width==1 || part_width==2 || part_width==4);
1028
1029 /* mv_cache
1030 B . . A T T T T
1031 U . . L . . , .
1032 U . . L . . . .
1033 U . . L . . , .
1034 . . . L . . . .
1035 */
1036
1037 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1038 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1039 tprintf("pred_motion match_count=%d\n", match_count);
1040 if(match_count > 1){ //most common
1041 *mx= mid_pred(A[0], B[0], C[0]);
1042 *my= mid_pred(A[1], B[1], C[1]);
1043 }else if(match_count==1){
1044 if(left_ref==ref){
1045 *mx= A[0];
1046 *my= A[1];
1047 }else if(top_ref==ref){
1048 *mx= B[0];
1049 *my= B[1];
1050 }else{
1051 *mx= C[0];
1052 *my= C[1];
1053 }
1054 }else{
1055 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1056 *mx= A[0];
1057 *my= A[1];
1058 }else{
1059 *mx= mid_pred(A[0], B[0], C[0]);
1060 *my= mid_pred(A[1], B[1], C[1]);
1061 }
1062 }
1063
1064 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);
1065 }
1066
1067 /**
1068 * gets the directionally predicted 16x8 MV.
1069 * @param n the block index
1070 * @param mx the x component of the predicted motion vector
1071 * @param my the y component of the predicted motion vector
1072 */
1073 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1074 if(n==0){
1075 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1076 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1077
1078 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);
1079
1080 if(top_ref == ref){
1081 *mx= B[0];
1082 *my= B[1];
1083 return;
1084 }
1085 }else{
1086 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1087 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1088
1089 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);
1090
1091 if(left_ref == ref){
1092 *mx= A[0];
1093 *my= A[1];
1094 return;
1095 }
1096 }
1097
1098 //RARE
1099 pred_motion(h, n, 4, list, ref, mx, my);
1100 }
1101
1102 /**
1103 * gets the directionally predicted 8x16 MV.
1104 * @param n the block index
1105 * @param mx the x component of the predicted motion vector
1106 * @param my the y component of the predicted motion vector
1107 */
1108 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1109 if(n==0){
1110 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1111 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
1112
1113 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);
1114
1115 if(left_ref == ref){
1116 *mx= A[0];
1117 *my= A[1];
1118 return;
1119 }
1120 }else{
1121 const int16_t * C;
1122 int diagonal_ref;
1123
1124 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1125
1126 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);
1127
1128 if(diagonal_ref == ref){
1129 *mx= C[0];
1130 *my= C[1];
1131 return;
1132 }
1133 }
1134
1135 //RARE
1136 pred_motion(h, n, 2, list, ref, mx, my);
1137 }
1138
1139 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1140 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1141 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1142
1143 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1144
1145 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1146 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1147 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1148
1149 *mx = *my = 0;
1150 return;
1151 }
1152
1153 pred_motion(h, 0, 4, 0, 0, mx, my);
1154
1155 return;
1156 }
1157
1158 static inline void direct_dist_scale_factor(H264Context * const h){
1159 const int poc = h->s.current_picture_ptr->poc;
1160 const int poc1 = h->ref_list[1][0].poc;
1161 int i;
1162 for(i=0; i<h->ref_count[0]; i++){
1163 int poc0 = h->ref_list[0][i].poc;
1164 int td = clip(poc1 - poc0, -128, 127);
1165 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1166 h->dist_scale_factor[i] = 256;
1167 }else{
1168 int tb = clip(poc - poc0, -128, 127);
1169 int tx = (16384 + (ABS(td) >> 1)) / td;
1170 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1171 }
1172 }
1173 }
1174 static inline void direct_ref_list_init(H264Context * const h){
1175 MpegEncContext * const s = &h->s;
1176 Picture * const ref1 = &h->ref_list[1][0];
1177 Picture * const cur = s->current_picture_ptr;
1178 int list, i, j;
1179 if(cur->pict_type == I_TYPE)
1180 cur->ref_count[0] = 0;
1181 if(cur->pict_type != B_TYPE)
1182 cur->ref_count[1] = 0;
1183 for(list=0; list<2; list++){
1184 cur->ref_count[list] = h->ref_count[list];
1185 for(j=0; j<h->ref_count[list]; j++)
1186 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1187 }
1188 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1189 return;
1190 for(list=0; list<2; list++){
1191 for(i=0; i<ref1->ref_count[list]; i++){
1192 const int poc = ref1->ref_poc[list][i];
1193 h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1194 for(j=0; j<h->ref_count[list]; j++)
1195 if(h->ref_list[list][j].poc == poc){
1196 h->map_col_to_list0[list][i] = j;
1197 break;
1198 }
1199 }
1200 }
1201 }
1202
1203 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1204 MpegEncContext * const s = &h->s;
1205 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1206 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1207 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1208 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1209 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1210 const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1211 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1212 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1213 const int is_b8x8 = IS_8X8(*mb_type);
1214 int sub_mb_type;
1215 int i8, i4;
1216
1217 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1218 /* FIXME save sub mb types from previous frames (or derive from MVs)
1219 * so we know exactly what block size to use */
1220 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1221 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1222 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1223 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1224 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1225 }else{
1226 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1227 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1228 }
1229 if(!is_b8x8)
1230 *mb_type |= MB_TYPE_DIRECT2;
1231
1232 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);
1233
1234 if(h->direct_spatial_mv_pred){
1235 int ref[2];
1236 int mv[2][2];
1237 int list;
1238
1239 /* ref = min(neighbors) */
1240 for(list=0; list<2; list++){
1241 int refa = h->ref_cache[list][scan8[0] - 1];
1242 int refb = h->ref_cache[list][scan8[0] - 8];
1243 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1244 if(refc == -2)
1245 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1246 ref[list] = refa;
1247 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1248 ref[list] = refb;
1249 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1250 ref[list] = refc;
1251 if(ref[list] < 0)
1252 ref[list] = -1;
1253 }
1254
1255 if(ref[0] < 0 && ref[1] < 0){
1256 ref[0] = ref[1] = 0;
1257 mv[0][0] = mv[0][1] =
1258 mv[1][0] = mv[1][1] = 0;
1259 }else{
1260 for(list=0; list<2; list++){
1261 if(ref[list] >= 0)
1262 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1263 else
1264 mv[list][0] = mv[list][1] = 0;
1265 }
1266 }
1267
1268 if(ref[1] < 0){
1269 *mb_type &= ~MB_TYPE_P0L1;
1270 sub_mb_type &= ~MB_TYPE_P0L1;
1271 }else if(ref[0] < 0){
1272 *mb_type &= ~MB_TYPE_P0L0;
1273 sub_mb_type &= ~MB_TYPE_P0L0;
1274 }
1275
1276 if(IS_16X16(*mb_type)){
1277 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1278 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1279 if(!IS_INTRA(mb_type_col)
1280 && ( (l1ref0[0] == 0 && ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1)
1281 || (l1ref0[0] < 0 && l1ref1[0] == 0 && ABS(l1mv1[0][0]) <= 1 && ABS(l1mv1[0][1]) <= 1
1282 && (h->x264_build>33 || !h->x264_build)))){
1283 if(ref[0] > 0)
1284 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1285 else
1286 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1287 if(ref[1] > 0)
1288 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1289 else
1290 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1291 }else{
1292 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1293 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1294 }
1295 }else{
1296 for(i8=0; i8<4; i8++){
1297 const int x8 = i8&1;
1298 const int y8 = i8>>1;
1299
1300 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1301 continue;
1302 h->sub_mb_type[i8] = sub_mb_type;
1303
1304 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1305 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1306 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1307 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1308
1309 /* col_zero_flag */
1310 if(!IS_INTRA(mb_type_col) && ( l1ref0[x8 + y8*h->b8_stride] == 0
1311 || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1312 && (h->x264_build>33 || !h->x264_build)))){
1313 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1314 for(i4=0; i4<4; i4++){
1315 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1316 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1317 if(ref[0] == 0)
1318 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1319 if(ref[1] == 0)
1320 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1321 }
1322 }
1323 }
1324 }
1325 }
1326 }else{ /* direct temporal mv pred */
1327 if(IS_16X16(*mb_type)){
1328 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1329 if(IS_INTRA(mb_type_col)){
1330 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1331 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1332 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1333 }else{
1334 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1335 : h->map_col_to_list0[1][l1ref1[0]];
1336 const int dist_scale_factor = h->dist_scale_factor[ref0];
1337 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1338 int mv_l0[2];
1339 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1340 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1341 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1342 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1343 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);
1344 }
1345 }else{
1346 for(i8=0; i8<4; i8++){
1347 const int x8 = i8&1;
1348 const int y8 = i8>>1;
1349 int ref0, dist_scale_factor;
1350 const int16_t (*l1mv)[2]= l1mv0;
1351
1352 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1353 continue;
1354 h->sub_mb_type[i8] = sub_mb_type;
1355 if(IS_INTRA(mb_type_col)){
1356 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1357 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1358 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1359 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1360 continue;
1361 }
1362
1363 ref0 = l1ref0[x8 + y8*h->b8_stride];
1364 if(ref0 >= 0)
1365 ref0 = h->map_col_to_list0[0][ref0];
1366 else{
1367 ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1368 l1mv= l1mv1;
1369 }
1370 dist_scale_factor = h->dist_scale_factor[ref0];
1371
1372 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1373 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1374 for(i4=0; i4<4; i4++){
1375 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1376 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1377 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1378 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1379 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1380 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1381 }
1382 }
1383 }
1384 }
1385 }
1386
1387 static inline void write_back_motion(H264Context *h, int mb_type){
1388 MpegEncContext * const s = &h->s;
1389 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1390 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1391 int list;
1392
1393 for(list=0; list<2; list++){
1394 int y;
1395 if(!USES_LIST(mb_type, list)){
1396 if(1){ //FIXME skip or never read if mb_type doesn't use it
1397 for(y=0; y<4; y++){
1398 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1399 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1400 }
1401 if( h->pps.cabac ) {
1402 /* FIXME needed ? */
1403 for(y=0; y<4; y++){
1404 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1405 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1406 }
1407 }
1408 for(y=0; y<2; y++){
1409 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]=
1410 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= LIST_NOT_USED;
1411 }
1412 }
1413 continue;
1414 }
1415
1416 for(y=0; y<4; y++){
1417 *(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];
1418 *(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];
1419 }
1420 if( h->pps.cabac ) {
1421 for(y=0; y<4; y++){
1422 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1423 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1424 }
1425 }
1426 for(y=0; y<2; y++){
1427 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1428 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1429 }
1430 }
1431
1432 if(h->slice_type == B_TYPE && h->pps.cabac){
1433 if(IS_8X8(mb_type)){
1434 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1435 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1436 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1437 }
1438 }
1439 }
1440
1441 /**
1442 * Decodes a network abstraction layer unit.
1443 * @param consumed is the number of bytes used as input
1444 * @param length is the length of the array
1445 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1446 * @returns decoded bytes, might be src+1 if no escapes
1447 */
1448 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1449 int i, si, di;
1450 uint8_t *dst;
1451
1452 // src[0]&0x80; //forbidden bit
1453 h->nal_ref_idc= src[0]>>5;
1454 h->nal_unit_type= src[0]&0x1F;
1455
1456 src++; length--;
1457 #if 0
1458 for(i=0; i<length; i++)
1459 printf("%2X ", src[i]);
1460 #endif
1461 for(i=0; i+1<length; i+=2){
1462 if(src[i]) continue;
1463 if(i>0 && src[i-1]==0) i--;
1464 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1465 if(src[i+2]!=3){
1466 /* startcode, so we must be past the end */
1467 length=i;
1468 }
1469 break;
1470 }
1471 }
1472
1473 if(i>=length-1){ //no escaped 0
1474 *dst_length= length;
1475 *consumed= length+1; //+1 for the header
1476 return src;
1477 }
1478
1479 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1480 dst= h->rbsp_buffer;
1481
1482 //printf("decoding esc\n");
1483 si=di=0;
1484 while(si<length){
1485 //remove escapes (very rare 1:2^22)
1486 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1487 if(src[si+2]==3){ //escape
1488 dst[di++]= 0;
1489 dst[di++]= 0;
1490 si+=3;
1491 continue;
1492 }else //next start code
1493 break;
1494 }
1495
1496 dst[di++]= src[si++];
1497 }
1498
1499 *dst_length= di;
1500 *consumed= si + 1;//+1 for the header
1501 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1502 return dst;
1503 }
1504
1505 #if 0
1506 /**
1507 * @param src the data which should be escaped
1508 * @param dst the target buffer, dst+1 == src is allowed as a special case
1509 * @param length the length of the src data
1510 * @param dst_length the length of the dst array
1511 * @returns length of escaped data in bytes or -1 if an error occured
1512 */
1513 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1514 int i, escape_count, si, di;
1515 uint8_t *temp;
1516
1517 assert(length>=0);
1518 assert(dst_length>0);
1519
1520 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1521
1522 if(length==0) return 1;
1523
1524 escape_count= 0;
1525 for(i=0; i<length; i+=2){
1526 if(src[i]) continue;
1527 if(i>0 && src[i-1]==0)
1528 i--;
1529 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1530 escape_count++;
1531 i+=2;
1532 }
1533 }
1534
1535 if(escape_count==0){
1536 if(dst+1 != src)
1537 memcpy(dst+1, src, length);
1538 return length + 1;
1539 }
1540
1541 if(length + escape_count + 1> dst_length)
1542 return -1;
1543
1544 //this should be damn rare (hopefully)
1545
1546 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1547 temp= h->rbsp_buffer;
1548 //printf("encoding esc\n");
1549
1550 si= 0;
1551 di= 0;
1552 while(si < length){
1553 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1554 temp[di++]= 0; si++;
1555 temp[di++]= 0; si++;
1556 temp[di++]= 3;
1557 temp[di++]= src[si++];
1558 }
1559 else
1560 temp[di++]= src[si++];
1561 }
1562 memcpy(dst+1, temp, length+escape_count);
1563
1564 assert(di == length+escape_count);
1565
1566 return di + 1;
1567 }
1568
1569 /**
1570 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1571 */
1572 static void encode_rbsp_trailing(PutBitContext *pb){
1573 int length;
1574 put_bits(pb, 1, 1);
1575 length= (-put_bits_count(pb))&7;
1576 if(length) put_bits(pb, length, 0);
1577 }
1578 #endif
1579
1580 /**
1581 * identifies the exact end of the bitstream
1582 * @return the length of the trailing, or 0 if damaged
1583 */
1584 static int decode_rbsp_trailing(uint8_t *src){
1585 int v= *src;
1586 int r;
1587
1588 tprintf("rbsp trailing %X\n", v);
1589
1590 for(r=1; r<9; r++){
1591 if(v&1) return r;
1592 v>>=1;
1593 }
1594 return 0;
1595 }
1596
1597 /**
1598 * idct tranforms the 16 dc values and dequantize them.
1599 * @param qp quantization parameter
1600 */
1601 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1602 const int qmul= dequant_coeff[qp][0];
1603 #define stride 16
1604 int i;
1605 int temp[16]; //FIXME check if this is a good idea
1606 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1607 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1608
1609 //memset(block, 64, 2*256);
1610 //return;
1611 for(i=0; i<4; i++){
1612 const int offset= y_offset[i];
1613 const int z0= block[offset+stride*0] + block[offset+stride*4];
1614 const int z1= block[offset+stride*0] - block[offset+stride*4];
1615 const int z2= block[offset+stride*1] - block[offset+stride*5];
1616 const int z3= block[offset+stride*1] + block[offset+stride*5];
1617
1618 temp[4*i+0]= z0+z3;
1619 temp[4*i+1]= z1+z2;
1620 temp[4*i+2]= z1-z2;
1621 temp[4*i+3]= z0-z3;
1622 }
1623
1624 for(i=0; i<4; i++){
1625 const int offset= x_offset[i];
1626 const int z0= temp[4*0+i] + temp[4*2+i];
1627 const int z1= temp[4*0+i] - temp[4*2+i];
1628 const int z2= temp[4*1+i] - temp[4*3+i];
1629 const int z3= temp[4*1+i] + temp[4*3+i];
1630
1631 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1632 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1633 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1634 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1635 }
1636 }
1637
1638 #if 0
1639 /**
1640 * dct tranforms the 16 dc values.
1641 * @param qp quantization parameter ??? FIXME
1642 */
1643 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1644 // const int qmul= dequant_coeff[qp][0];
1645 int i;
1646 int temp[16]; //FIXME check if this is a good idea
1647 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1648 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1649
1650 for(i=0; i<4; i++){
1651 const int offset= y_offset[i];
1652 const int z0= block[offset+stride*0] + block[offset+stride*4];
1653 const int z1= block[offset+stride*0] - block[offset+stride*4];
1654 const int z2= block[offset+stride*1] - block[offset+stride*5];
1655 const int z3= block[offset+stride*1] + block[offset+stride*5];
1656
1657 temp[4*i+0]= z0+z3;
1658 temp[4*i+1]= z1+z2;
1659 temp[4*i+2]= z1-z2;
1660 temp[4*i+3]= z0-z3;
1661 }
1662
1663 for(i=0; i<4; i++){
1664 const int offset= x_offset[i];
1665 const int z0= temp[4*0+i] + temp[4*2+i];
1666 const int z1= temp[4*0+i] - temp[4*2+i];
1667 const int z2= temp[4*1+i] - temp[4*3+i];
1668 const int z3= temp[4*1+i] + temp[4*3+i];
1669
1670 block[stride*0 +offset]= (z0 + z3)>>1;
1671 block[stride*2 +offset]= (z1 + z2)>>1;
1672 block[stride*8 +offset]= (z1 - z2)>>1;
1673 block[stride*10+offset]= (z0 - z3)>>1;
1674 }
1675 }
1676 #endif
1677
1678 #undef xStride
1679 #undef stride
1680
1681 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1682 const int qmul= dequant_coeff[qp][0];
1683 const int stride= 16*2;
1684 const int xStride= 16;
1685 int a,b,c,d,e;
1686
1687 a= block[stride*0 + xStride*0];
1688 b= block[stride*0 + xStride*1];
1689 c= block[stride*1 + xStride*0];
1690 d= block[stride*1 + xStride*1];
1691
1692 e= a-b;
1693 a= a+b;
1694 b= c-d;
1695 c= c+d;
1696
1697 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1698 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1699 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1700 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1701 }
1702
1703 #if 0
1704 static void chroma_dc_dct_c(DCTELEM *block){
1705 const int stride= 16*2;
1706 const int xStride= 16;
1707 int a,b,c,d,e;
1708
1709 a= block[stride*0 + xStride*0];
1710 b= block[stride*0 + xStride*1];
1711 c= block[stride*1 + xStride*0];
1712 d= block[stride*1 + xStride*1];
1713
1714 e= a-b;
1715 a= a+b;
1716 b= c-d;
1717 c= c+d;
1718
1719 block[stride*0 + xStride*0]= (a+c);
1720 block[stride*0 + xStride*1]= (e+b);
1721 block[stride*1 + xStride*0]= (a-c);
1722 block[stride*1 + xStride*1]= (e-b);
1723 }
1724 #endif
1725
1726 /**
1727 * gets the chroma qp.
1728 */
1729 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1730
1731 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1732 }
1733
1734
1735 #if 0
1736 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1737 int i;
1738 //FIXME try int temp instead of block
1739
1740 for(i=0; i<4; i++){
1741 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1742 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1743 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1744 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1745 const int z0= d0 + d3;
1746 const int z3= d0 - d3;
1747 const int z1= d1 + d2;
1748 const int z2= d1 - d2;
1749
1750 block[0 + 4*i]= z0 + z1;
1751 block[1 + 4*i]= 2*z3 + z2;
1752 block[2 + 4*i]= z0 - z1;
1753 block[3 + 4*i]= z3 - 2*z2;
1754 }
1755
1756 for(i=0; i<4; i++){
1757 const int z0= block[0*4 + i] + block[3*4 + i];
1758 const int z3= block[0*4 + i] - block[3*4 + i];
1759 const int z1= block[1*4 + i] + block[2*4 + i];
1760 const int z2= block[1*4 + i] - block[2*4 + i];
1761
1762 block[0*4 + i]= z0 + z1;
1763 block[1*4 + i]= 2*z3 + z2;
1764 block[2*4 + i]= z0 - z1;
1765 block[3*4 + i]= z3 - 2*z2;
1766 }
1767 }
1768 #endif
1769
1770 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1771 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1772 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1773 int i;
1774 const int * const quant_table= quant_coeff[qscale];
1775 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1776 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1777 const unsigned int threshold2= (threshold1<<1);
1778 int last_non_zero;
1779
1780 if(seperate_dc){
1781 if(qscale<=18){
1782 //avoid overflows
1783 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1784 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1785 const unsigned int dc_threshold2= (dc_threshold1<<1);
1786
1787 int level= block[0]*quant_coeff[qscale+18][0];
1788 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1789 if(level>0){
1790 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1791 block[0]= level;
1792 }else{
1793 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1794 block[0]= -level;
1795 }
1796 // last_non_zero = i;
1797 }else{
1798 block[0]=0;
1799 }
1800 }else{
1801 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1802 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1803 const unsigned int dc_threshold2= (dc_threshold1<<1);
1804
1805 int level= block[0]*quant_table[0];
1806 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1807 if(level>0){
1808 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1809 block[0]= level;
1810 }else{
1811 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1812 block[0]= -level;
1813 }
1814 // last_non_zero = i;
1815 }else{
1816 block[0]=0;
1817 }
1818 }
1819 last_non_zero= 0;
1820 i=1;
1821 }else{
1822 last_non_zero= -1;
1823 i=0;
1824 }
1825
1826 for(; i<16; i++){
1827 const int j= scantable[i];
1828 int level= block[j]*quant_table[j];
1829
1830 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1831 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1832 if(((unsigned)(level+threshold1))>threshold2){
1833 if(level>0){
1834 level= (bias + level)>>QUANT_SHIFT;
1835 block[j]= level;
1836 }else{
1837 level= (bias - level)>>QUANT_SHIFT;
1838 block[j]= -level;
1839 }
1840 last_non_zero = i;
1841 }else{
1842 block[j]=0;
1843 }
1844 }
1845
1846 return last_non_zero;
1847 }
1848
1849 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1850 const uint32_t a= ((uint32_t*)(src-stride))[0];
1851 ((uint32_t*)(src+0*stride))[0]= a;
1852 ((uint32_t*)(src+1*stride))[0]= a;
1853 ((uint32_t*)(src+2*stride))[0]= a;
1854 ((uint32_t*)(src+3*stride))[0]= a;
1855 }
1856
1857 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1858 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1859 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1860 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1861 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1862 }
1863
1864 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1865 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1866 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1867
1868 ((uint32_t*)(src+0*stride))[0]=
1869 ((uint32_t*)(src+1*stride))[0]=
1870 ((uint32_t*)(src+2*stride))[0]=
1871 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1872 }
1873
1874 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1875 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1876
1877 ((uint32_t*)(src+0*stride))[0]=
1878 ((uint32_t*)(src+1*stride))[0]=
1879 ((uint32_t*)(src+2*stride))[0]=
1880 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1881 }
1882
1883 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1884 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1885
1886 ((uint32_t*)(src+0*stride))[0]=
1887 ((uint32_t*)(src+1*stride))[0]=
1888 ((uint32_t*)(src+2*stride))[0]=
1889 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1890 }
1891
1892 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1893 ((uint32_t*)(src+0*stride))[0]=
1894 ((uint32_t*)(src+1*stride))[0]=
1895 ((uint32_t*)(src+2*stride))[0]=
1896 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1897 }
1898
1899
1900 #define LOAD_TOP_RIGHT_EDGE\
1901 const int t4= topright[0];\
1902 const int t5= topright[1];\
1903 const int t6= topright[2];\
1904 const int t7= topright[3];\
1905
1906 #define LOAD_LEFT_EDGE\
1907 const int l0= src[-1+0*stride];\
1908 const int l1= src[-1+1*stride];\
1909 const int l2= src[-1+2*stride];\
1910 const int l3= src[-1+3*stride];\
1911
1912 #define LOAD_TOP_EDGE\
1913 const int t0= src[ 0-1*stride];\
1914 const int t1= src[ 1-1*stride];\
1915 const int t2= src[ 2-1*stride];\
1916 const int t3= src[ 3-1*stride];\
1917
1918 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1919 const int lt= src[-1-1*stride];
1920 LOAD_TOP_EDGE
1921 LOAD_LEFT_EDGE
1922
1923 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1924 src[0+2*stride]=
1925 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1926 src[0+1*stride]=
1927 src[1+2*stride]=
1928 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1929 src[0+0*stride]=
1930 src[1+1*stride]=
1931 src[2+2*stride]=
1932 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1933 src[1+0*stride]=
1934 src[2+1*stride]=
1935 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1936 src[2+0*stride]=
1937 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1938 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1939 }
1940
1941 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1942 LOAD_TOP_EDGE
1943 LOAD_TOP_RIGHT_EDGE
1944 // LOAD_LEFT_EDGE
1945
1946 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1947 src[1+0*stride]=
1948 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1949 src[2+0*stride]=
1950 src[1+1*stride]=
1951 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1952 src[3+0*stride]=
1953 src[2+1*stride]=
1954 src[1+2*stride]=
1955 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1956 src[3+1*stride]=
1957 src[2+2*stride]=
1958 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1959 src[3+2*stride]=
1960 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1961 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1962 }
1963
1964 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1965 const int lt= src[-1-1*stride];
1966 LOAD_TOP_EDGE
1967 LOAD_LEFT_EDGE
1968 const __attribute__((unused)) int unu= l3;
1969
1970 src[0+0*stride]=
1971 src[1+2*stride]=(lt + t0 + 1)>>1;
1972 src[1+0*stride]=
1973 src[2+2*stride]=(t0 + t1 + 1)>>1;
1974 src[2+0*stride]=
1975 src[3+2*stride]=(t1 + t2 + 1)>>1;
1976 src[3+0*stride]=(t2 + t3 + 1)>>1;
1977 src[0+1*stride]=
1978 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1979 src[1+1*stride]=
1980 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1981 src[2+1*stride]=
1982 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1983 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1984 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1985 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1986 }
1987
1988 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1989 LOAD_TOP_EDGE
1990 LOAD_TOP_RIGHT_EDGE
1991 const __attribute__((unused)) int unu= t7;
1992
1993 src[0+0*stride]=(t0 + t1 + 1)>>1;
1994 src[1+0*stride]=
1995 src[0+2*stride]=(t1 + t2 + 1)>>1;
1996 src[2+0*stride]=
1997 src[1+2*stride]=(t2 + t3 + 1)>>1;
1998 src[3+0*stride]=
1999 src[2+2*stride]=(t3 + t4+ 1)>>1;
2000 src[3+2*stride]=(t4 + t5+ 1)>>1;
2001 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2002 src[1+1*stride]=
2003 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2004 src[2+1*stride]=
2005 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2006 src[3+1*stride]=
2007 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2008 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
2009 }
2010
2011 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
2012 LOAD_LEFT_EDGE
2013
2014 src[0+0*stride]=(l0 + l1 + 1)>>1;
2015 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2016 src[2+0*stride]=
2017 src[0+1*stride]=(l1 + l2 + 1)>>1;
2018 src[3+0*stride]=
2019 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2020 src[2+1*stride]=
2021 src[0+2*stride]=(l2 + l3 + 1)>>1;
2022 src[3+1*stride]=
2023 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2024 src[3+2*stride]=
2025 src[1+3*stride]=
2026 src[0+3*stride]=
2027 src[2+2*stride]=
2028 src[2+3*stride]=
2029 src[3+3*stride]=l3;
2030 }
2031
2032 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2033 const int lt= src[-1-1*stride];
2034 LOAD_TOP_EDGE
2035 LOAD_LEFT_EDGE
2036 const __attribute__((unused)) int unu= t3;
2037
2038 src[0+0*stride]=
2039 src[2+1*stride]=(lt + l0 + 1)>>1;
2040 src[1+0*stride]=
2041 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2042 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2043 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2044 src[0+1*stride]=
2045 src[2+2*stride]=(l0 + l1 + 1)>>1;
2046 src[1+1*stride]=
2047 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2048 src[0+2*stride]=
2049 src[2+3*stride]=(l1 + l2+ 1)>>1;
2050 src[1+2*stride]=
2051 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2052 src[0+3*stride]=(l2 + l3 + 1)>>1;
2053 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2054 }
2055
2056 static void pred16x16_vertical_c(uint8_t *src, int stride){
2057 int i;
2058 const uint32_t a= ((uint32_t*)(src-stride))[0];
2059 const uint32_t b= ((uint32_t*)(src-stride))[1];
2060 const uint32_t c= ((uint32_t*)(src-stride))[2];
2061 const uint32_t d= ((uint32_t*)(src-stride))[3];
2062
2063 for(i=0; i<16; i++){
2064 ((uint32_t*)(src+i*stride))[0]= a;
2065 ((uint32_t*)(src+i*stride))[1]= b;
2066 ((uint32_t*)(src+i*stride))[2]= c;
2067 ((uint32_t*)(src+i*stride))[3]= d;
2068 }
2069 }
2070
2071 static void pred16x16_horizontal_c(uint8_t *src, int stride){
2072 int i;
2073
2074 for(i=0; i<16; i++){
2075 ((uint32_t*)(src+i*stride))[0]=
2076 ((uint32_t*)(src+i*stride))[1]=
2077 ((uint32_t*)(src+i*stride))[2]=
2078 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2079 }
2080 }
2081
2082 static void pred16x16_dc_c(uint8_t *src, int stride){
2083 int i, dc=0;
2084
2085 for(i=0;i<16; i++){
2086 dc+= src[-1+i*stride];
2087 }
2088
2089 for(i=0;i<16; i++){
2090 dc+= src[i-stride];
2091 }
2092
2093 dc= 0x01010101*((dc + 16)>>5);
2094
2095 for(i=0; i<16; i++){
2096 ((uint32_t*)(src+i*stride))[0]=
2097 ((uint32_t*)(src+i*stride))[1]=
2098 ((uint32_t*)(src+i*stride))[2]=
2099 ((uint32_t*)(src+i*stride))[3]= dc;
2100 }
2101 }
2102
2103 static void pred16x16_left_dc_c(uint8_t *src, int stride){
2104 int i, dc=0;
2105
2106 for(i=0;i<16; i++){
2107 dc+= src[-1+i*stride];
2108 }
2109
2110 dc= 0x01010101*((dc + 8)>>4);
2111
2112 for(i=0; i<16; i++){
2113 ((uint32_t*)(src+i*stride))[0]=
2114 ((uint32_t*)(src+i*stride))[1]=
2115 ((uint32_t*)(src+i*stride))[2]=
2116 ((uint32_t*)(src+i*stride))[3]= dc;
2117 }
2118 }
2119
2120 static void pred16x16_top_dc_c(uint8_t *src, int stride){
2121 int i, dc=0;
2122
2123 for(i=0;i<16; i++){
2124 dc+= src[i-stride];
2125 }
2126 dc= 0x01010101*((dc + 8)>>4);
2127
2128 for(i=0; i<16; i++){
2129 ((uint32_t*)(src+i*stride))[0]=
2130 ((uint32_t*)(src+i*stride))[1]=
2131 ((uint32_t*)(src+i*stride))[2]=
2132 ((uint32_t*)(src+i*stride))[3]= dc;
2133 }
2134 }
2135
2136 static void pred16x16_128_dc_c(uint8_t *src, int stride){
2137 int i;
2138
2139 for(i=0; i<16; i++){
2140 ((uint32_t*)(src+i*stride))[0]=
2141 ((uint32_t*)(src+i*stride))[1]=
2142 ((uint32_t*)(src+i*stride))[2]=
2143 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2144 }
2145 }
2146
2147 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2148 int i, j, k;
2149 int a;
2150 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2151 const uint8_t * const src0 = src+7-stride;
2152 const uint8_t *src1 = src+8*stride-1;
2153 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2154 int H = src0[1] - src0[-1];
2155 int V = src1[0] - src2[ 0];
2156 for(k=2; k<=8; ++k) {
2157 src1 += stride; src2 -= stride;
2158 H += k*(src0[k] - src0[-k]);
2159 V += k*(src1[0] - src2[ 0]);
2160 }
2161 if(svq3){
2162 H = ( 5*(H/4) ) / 16;
2163 V = ( 5*(V/4) ) / 16;
2164
2165 /* required for 100% accuracy */
2166 i = H; H = V; V = i;
2167 }else{
2168 H = ( 5*H+32 ) >> 6;
2169 V = ( 5*V+32 ) >> 6;
2170 }
2171
2172 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2173 for(j=16; j>0; --j) {
2174 int b = a;
2175 a += V;
2176 for(i=-16; i<0; i+=4) {
2177 src[16+i] = cm[ (b ) >> 5 ];
2178 src[17+i] = cm[ (b+ H) >> 5 ];
2179 src[18+i] = cm[ (b+2*H) >> 5 ];
2180 src[19+i] = cm[ (b+3*H) >> 5 ];
2181 b += 4*H;
2182 }
2183 src += stride;
2184 }
2185 }
2186
2187 static void pred16x16_plane_c(uint8_t *src, int stride){
2188 pred16x16_plane_compat_c(src, stride, 0);
2189 }
2190
2191 static void pred8x8_vertical_c(uint8_t *src, int stride){
2192 int i;
2193 const uint32_t a= ((uint32_t*)(src-stride))[0];
2194 const uint32_t b= ((uint32_t*)(src-stride))[1];
2195
2196 for(i=0; i<8; i++){
2197 ((uint32_t*)(src+i*stride))[0]= a;
2198 ((uint32_t*)(src+i*stride))[1]= b;
2199 }
2200 }
2201
2202 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2203 int i;
2204
2205 for(i=0; i<8; i++){
2206 ((uint32_t*)(src+i*stride))[0]=
2207 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2208 }
2209 }
2210
2211 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2212 int i;
2213
2214 for(i=0; i<8; i++){
2215 ((uint32_t*)(src+i*stride))[0]=
2216 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2217 }
2218 }
2219
2220 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2221 int i;
2222 int dc0, dc2;
2223
2224 dc0=dc2=0;
2225 for(i=0;i<4; i++){
2226 dc0+= src[-1+i*stride];
2227 dc2+= src[-1+(i+4)*stride];
2228 }
2229 dc0= 0x01010101*((dc0 + 2)>>2);
2230 dc2= 0x01010101*((dc2 + 2)>>2);
2231
2232 for(i=0; i<4; i++){
2233 ((uint32_t*)(src+i*stride))[0]=
2234 ((uint32_t*)(src+i*stride))[1]= dc0;
2235 }
2236 for(i=4; i<8; i++){
2237 ((uint32_t*)(src+i*stride))[0]=
2238 ((uint32_t*)(src+i*stride))[1]= dc2;
2239 }
2240 }
2241
2242 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2243 int i;
2244 int dc0, dc1;
2245
2246 dc0=dc1=0;
2247 for(i=0;i<4; i++){
2248 dc0+= src[i-stride];
2249 dc1+= src[4+i-stride];
2250 }
2251 dc0= 0x01010101*((dc0 + 2)>>2);
2252 dc1= 0x01010101*((dc1 + 2)>>2);
2253
2254 for(i=0; i<4; i++){
2255 ((uint32_t*)(src+i*stride))[0]= dc0;
2256 ((uint32_t*)(src+i*stride))[1]= dc1;
2257 }
2258 for(i=4; i<8; i++){
2259 ((uint32_t*)(src+i*stride))[0]= dc0;
2260 ((uint32_t*)(src+i*stride))[1]= dc1;
2261 }
2262 }
2263
2264
2265 static void pred8x8_dc_c(uint8_t *src, int stride){
2266 int i;
2267 int dc0, dc1, dc2, dc3;
2268
2269 dc0=dc1=dc2=0;
2270 for(i=0;i<4; i++){
2271 dc0+= src[-1+i*stride] + src[i-stride];
2272 dc1+= src[4+i-stride];
2273 dc2+= src[-1+(i+4)*stride];
2274 }
2275 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2276 dc0= 0x01010101*((dc0 + 4)>>3);
2277 dc1= 0x01010101*((dc1 + 2)>>2);
2278 dc2= 0x01010101*((dc2 + 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]= dc2;
2286 ((uint32_t*)(src+i*stride))[1]= dc3;
2287 }
2288 }
2289
2290 static void pred8x8_plane_c(uint8_t *src, int stride){
2291 int j, k;
2292 int a;
2293 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2294 const uint8_t * const src0 = src+3-stride;
2295 const uint8_t *src1 = src+4*stride-1;
2296 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2297 int H = src0[1] - src0[-1];
2298 int V = src1[0] - src2[ 0];
2299 for(k=2; k<=4; ++k) {
2300 src1 += stride; src2 -= stride;
2301 H += k*(src0[k] - src0[-k]);
2302 V += k*(src1[0] - src2[ 0]);
2303 }
2304 H = ( 17*H+16 ) >> 5;
2305 V = ( 17*V+16 ) >> 5;
2306
2307 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2308 for(j=8; j>0; --j) {
2309 int b = a;
2310 a += V;
2311 src[0] = cm[ (b ) >> 5 ];
2312 src[1] = cm[ (b+ H) >> 5 ];
2313 src[2] = cm[ (b+2*H) >> 5 ];
2314 src[3] = cm[ (b+3*H) >> 5 ];
2315 src[4] = cm[ (b+4*H) >> 5 ];
2316 src[5] = cm[ (b+5*H) >> 5 ];
2317 src[6] = cm[ (b+6*H) >> 5 ];
2318 src[7] = cm[ (b+7*H) >> 5 ];
2319 src += stride;
2320 }
2321 }
2322
2323 #define SRC(x,y) src[(x)+(y)*stride]
2324 #define PL(y) \
2325 const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2326 #define PREDICT_8x8_LOAD_LEFT \
2327 const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2328 + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2329 PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2330 const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2331
2332 #define PT(x) \
2333 const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2334 #define PREDICT_8x8_LOAD_TOP \
2335 const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2336 + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2337 PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2338 const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2339 + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2340
2341 #define PTR(x) \
2342 t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2343 #define PREDICT_8x8_LOAD_TOPRIGHT \
2344 int t8, t9, t10, t11, t12, t13, t14, t15; \
2345 if(has_topright) { \
2346 PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2347 t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2348 } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2349
2350 #define PREDICT_8x8_LOAD_TOPLEFT \
2351 const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2352
2353 #define PREDICT_8x8_DC(v) \
2354 int y; \
2355 for( y = 0; y < 8; y++ ) { \
2356 ((uint32_t*)src)[0] = \
2357 ((uint32_t*)src)[1] = v; \
2358 src += stride; \
2359 }
2360
2361 static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2362 {
2363 PREDICT_8x8_DC(0x80808080);
2364 }
2365 static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2366 {
2367 PREDICT_8x8_LOAD_LEFT;
2368 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2369 PREDICT_8x8_DC(dc);
2370 }
2371 static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2372 {
2373 PREDICT_8x8_LOAD_TOP;
2374 const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2375 PREDICT_8x8_DC(dc);
2376 }
2377 static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2378 {
2379 PREDICT_8x8_LOAD_LEFT;
2380 PREDICT_8x8_LOAD_TOP;
2381 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2382 +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2383 PREDICT_8x8_DC(dc);
2384 }
2385 static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2386 {
2387 PREDICT_8x8_LOAD_LEFT;
2388 #define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2389 ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2390 ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2391 #undef ROW
2392 }
2393 static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2394 {
2395 int y;
2396 PREDICT_8x8_LOAD_TOP;
2397 src[0] = t0;
2398 src[1] = t1;
2399 src[2] = t2;
2400 src[3] = t3;
2401 src[4] = t4;
2402 src[5] = t5;
2403 src[6] = t6;
2404 src[7] = t7;
2405 for( y = 1; y < 8; y++ )
2406 *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2407 }
2408 static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2409 {
2410 PREDICT_8x8_LOAD_TOP;
2411 PREDICT_8x8_LOAD_TOPRIGHT;
2412 SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2413 SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2414 SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2415 SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2416 SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2417 SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2418 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;
2419 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;
2420 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;
2421 SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2422 SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2423 SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2424 SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2425 SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2426 SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2427 }
2428 static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2429 {
2430 PREDICT_8x8_LOAD_TOP;
2431 PREDICT_8x8_LOAD_LEFT;
2432 PREDICT_8x8_LOAD_TOPLEFT;
2433 SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2434 SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2435 SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2436 SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2437 SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2438 SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2439 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;
2440 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;
2441 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;
2442 SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2443 SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2444 SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2445 SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2446 SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2447 SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2448
2449 }
2450 static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2451 {
2452 PREDICT_8x8_LOAD_TOP;
2453 PREDICT_8x8_LOAD_LEFT;
2454 PREDICT_8x8_LOAD_TOPLEFT;
2455 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2456 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2457 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2458 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2459 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2460 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2461 SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2462 SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2463 SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2464 SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2465 SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2466 SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2467 SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2468 SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2469 SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2470 SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2471 SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2472 SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2473 SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2474 SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2475 SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2476 SRC(7,0)= (t6 + t7 + 1) >> 1;
2477 }
2478 static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2479 {
2480 PREDICT_8x8_LOAD_TOP;
2481 PREDICT_8x8_LOAD_LEFT;
2482 PREDICT_8x8_LOAD_TOPLEFT;
2483 SRC(0,7)= (l6 + l7 + 1) >> 1;
2484 SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2485 SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2486 SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2487 SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2488 SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2489 SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2490 SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2491 SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2492 SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2493 SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2494 SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2495 SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2496 SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2497 SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2498 SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2499 SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2500 SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2501 SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2502 SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2503 SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2504 SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2505 }
2506 static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2507 {
2508 PREDICT_8x8_LOAD_TOP;
2509 PREDICT_8x8_LOAD_TOPRIGHT;
2510 SRC(0,0)= (t0 + t1 + 1) >> 1;
2511 SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2512 SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2513 SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2514 SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2515 SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2516 SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2517 SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2518 SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2519 SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2520 SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2521 SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2522 SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2523 SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2524 SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2525 SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2526 SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2527 SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2528 SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2529 SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2530 SRC(7,6)= (t10 + t11 + 1) >> 1;
2531 SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2532 }
2533 static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2534 {
2535 PREDICT_8x8_LOAD_LEFT;
2536 SRC(0,0)= (l0 + l1 + 1) >> 1;
2537 SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2538 SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2539 SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2540 SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2541 SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2542 SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2543 SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2544 SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2545 SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2546 SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2547 SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2548 SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2549 SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2550 SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2551 SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2552 SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2553 SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2554 }
2555 #undef PREDICT_8x8_LOAD_LEFT
2556 #undef PREDICT_8x8_LOAD_TOP
2557 #undef PREDICT_8x8_LOAD_TOPLEFT
2558 #undef PREDICT_8x8_LOAD_TOPRIGHT
2559 #undef PREDICT_8x8_DC
2560 #undef PTR
2561 #undef PT
2562 #undef PL
2563 #undef SRC
2564
2565 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2566 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2567 int src_x_offset, int src_y_offset,
2568 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2569 MpegEncContext * const s = &h->s;
2570 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2571 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2572 const int luma_xy= (mx&3) + ((my&3)<<2);
2573 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2574 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2575 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2576 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2577 int extra_height= extra_width;
2578 int emu=0;
2579 const int full_mx= mx>>2;
2580 const int full_my= my>>2;
2581 const int pic_width = 16*s->mb_width;
2582 const int pic_height = 16*s->mb_height;
2583
2584 assert(pic->data[0]);
2585
2586 if(mx&7) extra_width -= 3;
2587 if(my&7) extra_height -= 3;
2588
2589 if( full_mx < 0-extra_width
2590 || full_my < 0-extra_height
2591 || full_mx + 16/*FIXME*/ > pic_width + extra_width
2592 || full_my + 16/*FIXME*/ > pic_height + extra_height){
2593 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);
2594 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2595 emu=1;
2596 }
2597
2598 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2599 if(!square){
2600 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2601 }
2602
2603 if(s->flags&CODEC_FLAG_GRAY) return;
2604
2605 if(emu){
2606 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);
2607 src_cb= s->edge_emu_buffer;
2608 }
2609 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2610
2611 if(emu){
2612 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);
2613 src_cr= s->edge_emu_buffer;
2614 }
2615 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2616 }
2617
2618 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2619 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2620 int x_offset, int y_offset,
2621 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2622 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2623 int list0, int list1){
2624 MpegEncContext * const s = &h->s;
2625 qpel_mc_func *qpix_op= qpix_put;
2626 h264_chroma_mc_func chroma_op= chroma_put;
2627
2628 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2629 dest_cb += x_offset + y_offset*s->uvlinesize;
2630 dest_cr += x_offset + y_offset*s->uvlinesize;
2631 x_offset += 8*s->mb_x;
2632 y_offset += 8*s->mb_y;
2633
2634 if(list0){
2635 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2636 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2637 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2638 qpix_op, chroma_op);
2639
2640 qpix_op= qpix_avg;
2641 chroma_op= chroma_avg;
2642 }
2643
2644 if(list1){
2645 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2646 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2647 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2648 qpix_op, chroma_op);
2649 }
2650 }
2651
2652 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2653 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2654 int x_offset, int y_offset,
2655 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2656 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2657 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2658 int list0, int list1){
2659 MpegEncContext * const s = &h->s;
2660
2661 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2662 dest_cb += x_offset + y_offset*s->uvlinesize;
2663 dest_cr += x_offset + y_offset*s->uvlinesize;
2664 x_offset += 8*s->mb_x;
2665 y_offset += 8*s->mb_y;
2666
2667 if(list0 && list1){
2668 /* don't optimize for luma-only case, since B-frames usually
2669 * use implicit weights => chroma too. */
2670 uint8_t *tmp_cb = s->obmc_scratchpad;
2671 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2672 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2673 int refn0 = h->ref_cache[0][ scan8[n] ];
2674 int refn1 = h->ref_cache[1][ scan8[n] ];
2675
2676 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2677 dest_y, dest_cb, dest_cr,
2678 x_offset, y_offset, qpix_put, chroma_put);
2679 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2680 tmp_y, tmp_cb, tmp_cr,
2681 x_offset, y_offset, qpix_put, chroma_put);
2682
2683 if(h->use_weight == 2){
2684 int weight0 = h->implicit_weight[refn0][refn1];
2685 int weight1 = 64 - weight0;
2686 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0, 0);
2687 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2688 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2689 }else{
2690 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2691 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2692 h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2693 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2694 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2695 h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2696 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2697 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2698 h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2699 }
2700 }else{
2701 int list = list1 ? 1 : 0;
2702 int refn = h->ref_cache[list][ scan8[n] ];
2703 Picture *ref= &h->ref_list[list][refn];
2704 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2705 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2706 qpix_put, chroma_put);
2707
2708 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2709 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2710 if(h->use_weight_chroma){
2711 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2712 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2713 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2714 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2715 }
2716 }
2717 }
2718
2719 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2720 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2721 int x_offset, int y_offset,
2722 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2723 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2724 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2725 int list0, int list1){
2726 if((h->use_weight==2 && list0 && list1
2727 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2728 || h->use_weight==1)
2729 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2730 x_offset, y_offset, qpix_put, chroma_put,
2731 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2732 else
2733 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2734 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2735 }
2736
2737 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2738 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2739 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2740 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2741 MpegEncContext * const s = &h->s;
2742 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2743 const int mb_type= s->current_picture.mb_type[mb_xy];
2744
2745 assert(IS_INTER(mb_type));
2746
2747 if(IS_16X16(mb_type)){
2748 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2749 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2750 &weight_op[0], &weight_avg[0],
2751 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2752 }else if(IS_16X8(mb_type)){
2753 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2754 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2755 &weight_op[1], &weight_avg[1],
2756 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2757 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2758 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2759 &weight_op[1], &weight_avg[1],
2760 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2761 }else if(IS_8X16(mb_type)){
2762 mc_part(h, 0, 0, 8, 8*