Mercurial > libavcodec.hg
annotate h264.c @ 1333:a1cc1810d58f libavcodec
build error on Alpha patch by (Sam Hocevar <sam at zoy dot org>)
author | michaelni |
---|---|
date | Sun, 29 Jun 2003 00:23:30 +0000 |
parents | c05c381a9c47 |
children | 6e5d4ec4f3ab |
rev | line source |
---|---|
1168 | 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 #undef NDEBUG | |
35 #include <assert.h> | |
36 | |
37 #define interlaced_dct interlaced_dct_is_a_bad_name | |
38 #define mb_intra mb_intra_isnt_initalized_see_mb_type | |
39 | |
40 #define LUMA_DC_BLOCK_INDEX 25 | |
41 #define CHROMA_DC_BLOCK_INDEX 26 | |
42 | |
43 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8 | |
44 #define COEFF_TOKEN_VLC_BITS 8 | |
45 #define TOTAL_ZEROS_VLC_BITS 9 | |
46 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3 | |
47 #define RUN_VLC_BITS 3 | |
48 #define RUN7_VLC_BITS 6 | |
49 | |
50 #define MAX_SPS_COUNT 32 | |
51 #define MAX_PPS_COUNT 256 | |
52 | |
53 #define MAX_MMCO_COUNT 66 | |
54 | |
55 /** | |
56 * Sequence parameter set | |
57 */ | |
58 typedef struct SPS{ | |
59 | |
60 int profile_idc; | |
61 int level_idc; | |
62 int multiple_slice_groups; ///< more_than_one_slice_group_allowed_flag | |
63 int arbitrary_slice_order; ///< arbitrary_slice_order_allowed_flag | |
64 int redundant_slices; ///< redundant_slices_allowed_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 required_frame_num_update_behaviour_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 vui_parameters_present_flag; | |
80 int sar_width; | |
81 int sar_height; | |
82 short offset_for_ref_frame[256]; //FIXME dyn aloc? | |
83 }SPS; | |
84 | |
85 /** | |
86 * Picture parameter set | |
87 */ | |
88 typedef struct PPS{ | |
89 int sps_id; | |
90 int cabac; ///< entropy_coding_mode_flag | |
91 int pic_order_present; ///< pic_order_present_flag | |
92 int slice_group_count; ///< num_slice_groups_minus1 + 1 | |
93 int mb_slice_group_map_type; | |
94 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1 | |
95 int weighted_pred; ///< weighted_pred_flag | |
96 int weighted_bipred_idc; | |
97 int init_qp; ///< pic_init_qp_minus26 + 26 | |
98 int init_qs; ///< pic_init_qs_minus26 + 26 | |
99 int chroma_qp_index_offset; | |
100 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag | |
101 int constrained_intra_pred; ///< constrained_intra_pred_flag | |
102 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag | |
103 int crop; ///< frame_cropping_flag | |
104 int crop_left; ///< frame_cropping_rect_left_offset | |
105 int crop_right; ///< frame_cropping_rect_right_offset | |
106 int crop_top; ///< frame_cropping_rect_top_offset | |
107 int crop_bottom; ///< frame_cropping_rect_bottom_offset | |
108 }PPS; | |
109 | |
110 /** | |
111 * Memory management control operation opcode. | |
112 */ | |
113 typedef enum MMCOOpcode{ | |
114 MMCO_END=0, | |
115 MMCO_SHORT2UNUSED, | |
116 MMCO_LONG2UNUSED, | |
117 MMCO_SHORT2LONG, | |
118 MMCO_SET_MAX_LONG, | |
119 MMCO_RESET, | |
120 MMCO_LONG, | |
121 } MMCOOpcode; | |
122 | |
123 /** | |
124 * Memory management control operation. | |
125 */ | |
126 typedef struct MMCO{ | |
127 MMCOOpcode opcode; | |
128 int short_frame_num; | |
129 int long_index; | |
130 } MMCO; | |
131 | |
132 /** | |
133 * H264Context | |
134 */ | |
135 typedef struct H264Context{ | |
136 MpegEncContext s; | |
137 int nal_ref_idc; | |
138 int nal_unit_type; | |
139 #define NAL_SLICE 1 | |
140 #define NAL_DPA 2 | |
141 #define NAL_DPB 3 | |
142 #define NAL_DPC 4 | |
143 #define NAL_IDR_SLICE 5 | |
144 #define NAL_SEI 6 | |
145 #define NAL_SPS 7 | |
146 #define NAL_PPS 8 | |
147 #define NAL_PICTURE_DELIMITER 9 | |
148 #define NAL_FILTER_DATA 10 | |
149 uint8_t *rbsp_buffer; | |
150 int rbsp_buffer_size; | |
151 | |
152 int chroma_qp; //QPc | |
153 | |
154 int prev_mb_skiped; //FIXME remove (IMHO not used) | |
155 | |
156 //prediction stuff | |
157 int chroma_pred_mode; | |
158 int intra16x16_pred_mode; | |
159 | |
160 int8_t intra4x4_pred_mode_cache[5*8]; | |
161 int8_t (*intra4x4_pred_mode)[8]; | |
162 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp? | |
163 void (*pred8x8 [4+3])(uint8_t *src, int stride); | |
164 void (*pred16x16[4+3])(uint8_t *src, int stride); | |
165 unsigned int topleft_samples_available; | |
166 unsigned int top_samples_available; | |
167 unsigned int topright_samples_available; | |
168 unsigned int left_samples_available; | |
169 | |
170 /** | |
171 * non zero coeff count cache. | |
172 * is 64 if not available. | |
173 */ | |
174 uint8_t non_zero_count_cache[6*8]; | |
175 uint8_t (*non_zero_count)[16]; | |
176 | |
177 /** | |
178 * Motion vector cache. | |
179 */ | |
180 int16_t mv_cache[2][5*8][2]; | |
181 int8_t ref_cache[2][5*8]; | |
182 #define LIST_NOT_USED -1 //FIXME rename? | |
183 #define PART_NOT_AVAILABLE -2 | |
184 | |
185 /** | |
186 * is 1 if the specific list MV&references are set to 0,0,-2. | |
187 */ | |
188 int mv_cache_clean[2]; | |
189 | |
190 int block_offset[16+8]; | |
191 int chroma_subblock_offset[16]; //FIXME remove | |
192 | |
193 uint16_t *mb2b_xy; //FIXME are these 4 a good idea? | |
194 uint16_t *mb2b8_xy; | |
195 int b_stride; | |
196 int b8_stride; | |
197 | |
1234 | 198 int halfpel_flag; |
199 int thirdpel_flag; | |
200 | |
1319 | 201 int unknown_svq3_flag; |
202 int next_slice_index; | |
203 | |
1168 | 204 SPS sps_buffer[MAX_SPS_COUNT]; |
205 SPS sps; ///< current sps | |
206 | |
207 PPS pps_buffer[MAX_PPS_COUNT]; | |
208 /** | |
209 * current pps | |
210 */ | |
211 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that? | |
212 | |
213 int slice_num; | |
214 uint8_t *slice_table_base; | |
215 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1 | |
216 int slice_type; | |
217 int slice_type_fixed; | |
218 | |
219 //interlacing specific flags | |
220 int mb_field_decoding_flag; | |
221 | |
222 int sub_mb_type[4]; | |
223 | |
224 //POC stuff | |
225 int poc_lsb; | |
226 int poc_msb; | |
227 int delta_poc_bottom; | |
228 int delta_poc[2]; | |
229 int frame_num; | |
230 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0 | |
231 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0 | |
232 int frame_num_offset; ///< for POC type 2 | |
233 int prev_frame_num_offset; ///< for POC type 2 | |
234 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2 | |
235 | |
236 /** | |
237 * frame_num for frames or 2*frame_num for field pics. | |
238 */ | |
239 int curr_pic_num; | |
240 | |
241 /** | |
242 * max_frame_num or 2*max_frame_num for field pics. | |
243 */ | |
244 int max_pic_num; | |
245 | |
246 //Weighted pred stuff | |
247 int luma_log2_weight_denom; | |
248 int chroma_log2_weight_denom; | |
249 int luma_weight[2][16]; | |
250 int luma_offset[2][16]; | |
251 int chroma_weight[2][16][2]; | |
252 int chroma_offset[2][16][2]; | |
253 | |
254 //deblock | |
255 int disable_deblocking_filter_idc; | |
256 int slice_alpha_c0_offset_div2; | |
257 int slice_beta_offset_div2; | |
258 | |
259 int redundant_pic_count; | |
260 | |
261 int direct_spatial_mv_pred; | |
262 | |
263 /** | |
264 * num_ref_idx_l0/1_active_minus1 + 1 | |
265 */ | |
266 int ref_count[2];// FIXME split for AFF | |
267 Picture *short_ref[16]; | |
268 Picture *long_ref[16]; | |
269 Picture default_ref_list[2][32]; | |
270 Picture ref_list[2][32]; //FIXME size? | |
271 Picture field_ref_list[2][32]; //FIXME size? | |
272 | |
273 /** | |
274 * memory management control operations buffer. | |
275 */ | |
276 MMCO mmco[MAX_MMCO_COUNT]; | |
277 int mmco_index; | |
278 | |
279 int long_ref_count; ///< number of actual long term references | |
280 int short_ref_count; ///< number of actual short term references | |
281 | |
282 //data partitioning | |
283 GetBitContext intra_gb; | |
284 GetBitContext inter_gb; | |
285 GetBitContext *intra_gb_ptr; | |
286 GetBitContext *inter_gb_ptr; | |
287 | |
288 DCTELEM mb[16*24] __align8; | |
289 }H264Context; | |
290 | |
291 static VLC coeff_token_vlc[4]; | |
292 static VLC chroma_dc_coeff_token_vlc; | |
293 | |
294 static VLC total_zeros_vlc[15]; | |
295 static VLC chroma_dc_total_zeros_vlc[3]; | |
296 | |
297 static VLC run_vlc[6]; | |
298 static VLC run7_vlc; | |
299 | |
1234 | 300 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp); |
301 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc); | |
302 | |
1269 | 303 static inline uint32_t pack16to32(int a, int b){ |
304 #ifdef WORDS_BIGENDIAN | |
305 return (b&0xFFFF) + (a<<16); | |
306 #else | |
307 return (a&0xFFFF) + (b<<16); | |
308 #endif | |
309 } | |
310 | |
1168 | 311 /** |
312 * fill a rectangle. | |
313 * @param h height of the recatangle, should be a constant | |
314 * @param w width of the recatangle, should be a constant | |
315 * @param size the size of val (1 or 4), should be a constant | |
316 */ | |
1187 | 317 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined |
318 uint8_t *p= (uint8_t*)vp; | |
1168 | 319 assert(size==1 || size==4); |
320 | |
321 w *= size; | |
322 stride *= size; | |
323 | |
324 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it | |
325 if(w==2 && h==2){ | |
326 *(uint16_t*)(p + 0)= | |
327 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101; | |
328 }else if(w==2 && h==4){ | |
329 *(uint16_t*)(p + 0*stride)= | |
330 *(uint16_t*)(p + 1*stride)= | |
331 *(uint16_t*)(p + 2*stride)= | |
332 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101; | |
1252 | 333 }else if(w==4 && h==1){ |
334 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101; | |
1168 | 335 }else if(w==4 && h==2){ |
336 *(uint32_t*)(p + 0*stride)= | |
337 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101; | |
338 }else if(w==4 && h==4){ | |
339 *(uint32_t*)(p + 0*stride)= | |
340 *(uint32_t*)(p + 1*stride)= | |
341 *(uint32_t*)(p + 2*stride)= | |
342 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101; | |
343 }else if(w==8 && h==1){ | |
344 *(uint32_t*)(p + 0)= | |
345 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101; | |
346 }else if(w==8 && h==2){ | |
347 *(uint32_t*)(p + 0 + 0*stride)= | |
348 *(uint32_t*)(p + 4 + 0*stride)= | |
349 *(uint32_t*)(p + 0 + 1*stride)= | |
350 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101; | |
351 }else if(w==8 && h==4){ | |
352 *(uint64_t*)(p + 0*stride)= | |
353 *(uint64_t*)(p + 1*stride)= | |
354 *(uint64_t*)(p + 2*stride)= | |
355 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL; | |
356 }else if(w==16 && h==2){ | |
357 *(uint64_t*)(p + 0+0*stride)= | |
358 *(uint64_t*)(p + 8+0*stride)= | |
359 *(uint64_t*)(p + 0+1*stride)= | |
360 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL; | |
361 }else if(w==16 && h==4){ | |
362 *(uint64_t*)(p + 0+0*stride)= | |
363 *(uint64_t*)(p + 8+0*stride)= | |
364 *(uint64_t*)(p + 0+1*stride)= | |
365 *(uint64_t*)(p + 8+1*stride)= | |
366 *(uint64_t*)(p + 0+2*stride)= | |
367 *(uint64_t*)(p + 8+2*stride)= | |
368 *(uint64_t*)(p + 0+3*stride)= | |
369 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL; | |
370 }else | |
371 assert(0); | |
372 } | |
373 | |
374 static inline void fill_caches(H264Context *h, int mb_type){ | |
375 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
376 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 377 int topleft_xy, top_xy, topright_xy, left_xy[2]; |
378 int topleft_type, top_type, topright_type, left_type[2]; | |
379 int left_block[4]; | |
380 int i; | |
381 | |
382 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it | |
383 | |
384 if(h->sps.mb_aff){ | |
385 //FIXME | |
386 }else{ | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
387 topleft_xy = mb_xy-1 - s->mb_stride; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
388 top_xy = mb_xy - s->mb_stride; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
389 topright_xy= mb_xy+1 - s->mb_stride; |
1168 | 390 left_xy[0] = mb_xy-1; |
391 left_xy[1] = mb_xy-1; | |
392 left_block[0]= 0; | |
393 left_block[1]= 1; | |
394 left_block[2]= 2; | |
395 left_block[3]= 3; | |
396 } | |
397 | |
398 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0; | |
399 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0; | |
400 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0; | |
401 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0; | |
402 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0; | |
403 | |
404 if(IS_INTRA(mb_type)){ | |
405 h->topleft_samples_available= | |
406 h->top_samples_available= | |
407 h->left_samples_available= 0xFFFF; | |
408 h->topright_samples_available= 0xEEEA; | |
409 | |
410 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){ | |
411 h->topleft_samples_available= 0xB3FF; | |
412 h->top_samples_available= 0x33FF; | |
413 h->topright_samples_available= 0x26EA; | |
414 } | |
415 for(i=0; i<2; i++){ | |
416 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){ | |
417 h->topleft_samples_available&= 0xDF5F; | |
418 h->left_samples_available&= 0x5F5F; | |
419 } | |
420 } | |
421 | |
422 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred)) | |
423 h->topleft_samples_available&= 0x7FFF; | |
424 | |
425 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred)) | |
426 h->topright_samples_available&= 0xFBFF; | |
427 | |
428 if(IS_INTRA4x4(mb_type)){ | |
429 if(IS_INTRA4x4(top_type)){ | |
430 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4]; | |
431 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5]; | |
432 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6]; | |
433 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3]; | |
434 }else{ | |
435 int pred; | |
436 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred)) | |
437 pred= 2; | |
438 else{ | |
439 pred= -1; | |
440 } | |
441 h->intra4x4_pred_mode_cache[4+8*0]= | |
442 h->intra4x4_pred_mode_cache[5+8*0]= | |
443 h->intra4x4_pred_mode_cache[6+8*0]= | |
444 h->intra4x4_pred_mode_cache[7+8*0]= pred; | |
445 } | |
446 for(i=0; i<2; i++){ | |
447 if(IS_INTRA4x4(left_type[i])){ | |
448 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]]; | |
449 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]]; | |
450 }else{ | |
451 int pred; | |
452 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred)) | |
453 pred= 2; | |
454 else{ | |
455 pred= -1; | |
456 } | |
457 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= | |
458 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred; | |
459 } | |
460 } | |
461 } | |
462 } | |
463 | |
464 | |
465 /* | |
466 0 . T T. T T T T | |
467 1 L . .L . . . . | |
468 2 L . .L . . . . | |
469 3 . T TL . . . . | |
470 4 L . .L . . . . | |
471 5 L . .. . . . . | |
472 */ | |
473 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec) | |
474 if(top_type){ | |
475 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0]; | |
476 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1]; | |
477 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2]; | |
478 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3]; | |
479 | |
480 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7]; | |
481 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8]; | |
482 | |
483 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10]; | |
484 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11]; | |
485 }else{ | |
486 h->non_zero_count_cache[4+8*0]= | |
487 h->non_zero_count_cache[5+8*0]= | |
488 h->non_zero_count_cache[6+8*0]= | |
489 h->non_zero_count_cache[7+8*0]= | |
490 | |
491 h->non_zero_count_cache[1+8*0]= | |
492 h->non_zero_count_cache[2+8*0]= | |
493 | |
494 h->non_zero_count_cache[1+8*3]= | |
495 h->non_zero_count_cache[2+8*3]= 64; | |
496 } | |
497 | |
498 if(left_type[0]){ | |
499 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6]; | |
500 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5]; | |
501 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block | |
502 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12]; | |
503 }else{ | |
504 h->non_zero_count_cache[3+8*1]= | |
505 h->non_zero_count_cache[3+8*2]= | |
506 h->non_zero_count_cache[0+8*1]= | |
507 h->non_zero_count_cache[0+8*4]= 64; | |
508 } | |
509 | |
510 if(left_type[1]){ | |
511 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4]; | |
512 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3]; | |
513 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8]; | |
514 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11]; | |
515 }else{ | |
516 h->non_zero_count_cache[3+8*3]= | |
517 h->non_zero_count_cache[3+8*4]= | |
518 h->non_zero_count_cache[0+8*2]= | |
519 h->non_zero_count_cache[0+8*5]= 64; | |
520 } | |
521 | |
522 #if 1 | |
523 if(IS_INTER(mb_type)){ | |
524 int list; | |
525 for(list=0; list<2; list++){ | |
526 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){ | |
527 /*if(!h->mv_cache_clean[list]){ | |
528 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all? | |
529 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t)); | |
530 h->mv_cache_clean[list]= 1; | |
531 }*/ | |
532 continue; //FIXME direct mode ... | |
533 } | |
534 h->mv_cache_clean[list]= 0; | |
535 | |
536 if(IS_INTER(topleft_type)){ | |
537 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride; | |
538 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride; | |
539 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy]; | |
540 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy]; | |
541 }else{ | |
542 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0; | |
543 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
544 } | |
545 | |
546 if(IS_INTER(top_type)){ | |
547 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride; | |
548 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride; | |
549 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0]; | |
550 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1]; | |
551 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2]; | |
552 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3]; | |
553 h->ref_cache[list][scan8[0] + 0 - 1*8]= | |
554 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0]; | |
555 h->ref_cache[list][scan8[0] + 2 - 1*8]= | |
556 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1]; | |
557 }else{ | |
558 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= | |
559 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= | |
560 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= | |
561 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0; | |
562 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101; | |
563 } | |
564 | |
565 if(IS_INTER(topright_type)){ | |
566 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride; | |
567 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride; | |
568 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy]; | |
569 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy]; | |
570 }else{ | |
571 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0; | |
572 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
573 } | |
574 | |
575 //FIXME unify cleanup or sth | |
576 if(IS_INTER(left_type[0])){ | |
577 const int b_xy= h->mb2b_xy[left_xy[0]] + 3; | |
578 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1; | |
579 *(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]]; | |
580 *(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]]; | |
581 h->ref_cache[list][scan8[0] - 1 + 0*8]= | |
582 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)]; | |
583 }else{ | |
584 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]= | |
585 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0; | |
586 h->ref_cache[list][scan8[0] - 1 + 0*8]= | |
587 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
588 } | |
589 | |
590 if(IS_INTER(left_type[1])){ | |
591 const int b_xy= h->mb2b_xy[left_xy[1]] + 3; | |
592 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1; | |
593 *(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]]; | |
594 *(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]]; | |
595 h->ref_cache[list][scan8[0] - 1 + 2*8]= | |
596 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)]; | |
597 }else{ | |
598 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]= | |
599 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0; | |
600 h->ref_cache[list][scan8[0] - 1 + 2*8]= | |
601 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
602 } | |
603 | |
604 h->ref_cache[list][scan8[5 ]+1] = | |
605 h->ref_cache[list][scan8[7 ]+1] = | |
606 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else) | |
607 h->ref_cache[list][scan8[4 ]] = | |
608 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE; | |
609 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]= | |
610 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]= | |
611 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else) | |
612 *(uint32_t*)h->mv_cache [list][scan8[4 ]]= | |
613 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0; | |
614 } | |
615 //FIXME | |
616 | |
617 } | |
618 #endif | |
619 } | |
620 | |
621 static inline void write_back_intra_pred_mode(H264Context *h){ | |
622 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
623 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 624 |
625 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1]; | |
626 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2]; | |
627 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3]; | |
628 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4]; | |
629 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4]; | |
630 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4]; | |
631 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4]; | |
632 } | |
633 | |
634 /** | |
635 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. | |
636 */ | |
637 static inline int check_intra4x4_pred_mode(H264Context *h){ | |
638 MpegEncContext * const s = &h->s; | |
639 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0}; | |
640 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED}; | |
641 int i; | |
642 | |
643 if(!(h->top_samples_available&0x8000)){ | |
644 for(i=0; i<4; i++){ | |
645 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ]; | |
646 if(status<0){ | |
647 fprintf(stderr, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y); | |
648 return -1; | |
649 } else if(status){ | |
650 h->intra4x4_pred_mode_cache[scan8[0] + i]= status; | |
651 } | |
652 } | |
653 } | |
654 | |
655 if(!(h->left_samples_available&0x8000)){ | |
656 for(i=0; i<4; i++){ | |
657 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ]; | |
658 if(status<0){ | |
659 fprintf(stderr, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y); | |
660 return -1; | |
661 } else if(status){ | |
662 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status; | |
663 } | |
664 } | |
665 } | |
666 | |
667 return 0; | |
668 } //FIXME cleanup like next | |
669 | |
670 /** | |
671 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. | |
672 */ | |
673 static inline int check_intra_pred_mode(H264Context *h, int mode){ | |
674 MpegEncContext * const s = &h->s; | |
675 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1}; | |
676 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8}; | |
677 | |
678 if(!(h->top_samples_available&0x8000)){ | |
679 mode= top[ mode ]; | |
680 if(mode<0){ | |
681 fprintf(stderr, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y); | |
682 return -1; | |
683 } | |
684 } | |
685 | |
686 if(!(h->left_samples_available&0x8000)){ | |
687 mode= left[ mode ]; | |
688 if(mode<0){ | |
689 fprintf(stderr, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y); | |
690 return -1; | |
691 } | |
692 } | |
693 | |
694 return mode; | |
695 } | |
696 | |
697 /** | |
698 * gets the predicted intra4x4 prediction mode. | |
699 */ | |
700 static inline int pred_intra_mode(H264Context *h, int n){ | |
701 const int index8= scan8[n]; | |
702 const int left= h->intra4x4_pred_mode_cache[index8 - 1]; | |
703 const int top = h->intra4x4_pred_mode_cache[index8 - 8]; | |
704 const int min= FFMIN(left, top); | |
705 | |
1170 | 706 tprintf("mode:%d %d min:%d\n", left ,top, min); |
1168 | 707 |
708 if(min<0) return DC_PRED; | |
709 else return min; | |
710 } | |
711 | |
712 static inline void write_back_non_zero_count(H264Context *h){ | |
713 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
714 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 715 |
716 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4]; | |
717 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4]; | |
718 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4]; | |
719 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4]; | |
720 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3]; | |
721 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2]; | |
722 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1]; | |
723 | |
724 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2]; | |
725 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2]; | |
726 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1]; | |
727 | |
728 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5]; | |
729 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5]; | |
730 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4]; | |
731 } | |
732 | |
733 /** | |
734 * gets the predicted number of non zero coefficients. | |
735 * @param n block index | |
736 */ | |
737 static inline int pred_non_zero_count(H264Context *h, int n){ | |
738 const int index8= scan8[n]; | |
739 const int left= h->non_zero_count_cache[index8 - 1]; | |
740 const int top = h->non_zero_count_cache[index8 - 8]; | |
741 int i= left + top; | |
742 | |
743 if(i<64) i= (i+1)>>1; | |
744 | |
1170 | 745 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31); |
1168 | 746 |
747 return i&31; | |
748 } | |
749 | |
1169 | 750 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){ |
751 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ]; | |
752 | |
753 if(topright_ref != PART_NOT_AVAILABLE){ | |
754 *C= h->mv_cache[list][ i - 8 + part_width ]; | |
755 return topright_ref; | |
756 }else{ | |
1170 | 757 tprintf("topright MV not available\n"); |
758 | |
1169 | 759 *C= h->mv_cache[list][ i - 8 - 1 ]; |
760 return h->ref_cache[list][ i - 8 - 1 ]; | |
761 } | |
762 } | |
763 | |
1168 | 764 /** |
765 * gets the predicted MV. | |
766 * @param n the block index | |
767 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4) | |
768 * @param mx the x component of the predicted motion vector | |
769 * @param my the y component of the predicted motion vector | |
770 */ | |
771 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){ | |
772 const int index8= scan8[n]; | |
773 const int top_ref= h->ref_cache[list][ index8 - 8 ]; | |
774 const int left_ref= h->ref_cache[list][ index8 - 1 ]; | |
775 const int16_t * const A= h->mv_cache[list][ index8 - 1 ]; | |
776 const int16_t * const B= h->mv_cache[list][ index8 - 8 ]; | |
1169 | 777 const int16_t * C; |
778 int diagonal_ref, match_count; | |
779 | |
1168 | 780 assert(part_width==1 || part_width==2 || part_width==4); |
1169 | 781 |
1168 | 782 /* mv_cache |
783 B . . A T T T T | |
784 U . . L . . , . | |
785 U . . L . . . . | |
786 U . . L . . , . | |
787 . . . L . . . . | |
788 */ | |
1169 | 789 |
790 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width); | |
791 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref); | |
792 | |
793 if(match_count > 1){ //most common | |
794 *mx= mid_pred(A[0], B[0], C[0]); | |
795 *my= mid_pred(A[1], B[1], C[1]); | |
796 }else if(match_count==1){ | |
797 if(left_ref==ref){ | |
798 *mx= A[0]; | |
799 *my= A[1]; | |
800 }else if(top_ref==ref){ | |
801 *mx= B[0]; | |
802 *my= B[1]; | |
803 }else{ | |
804 *mx= C[0]; | |
805 *my= C[1]; | |
806 } | |
807 }else{ | |
808 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){ | |
809 *mx= A[0]; | |
810 *my= A[1]; | |
1168 | 811 }else{ |
812 *mx= mid_pred(A[0], B[0], C[0]); | |
813 *my= mid_pred(A[1], B[1], C[1]); | |
814 } | |
1169 | 815 } |
1168 | 816 |
1187 | 817 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); |
1168 | 818 } |
819 | |
820 /** | |
821 * gets the directionally predicted 16x8 MV. | |
822 * @param n the block index | |
823 * @param mx the x component of the predicted motion vector | |
824 * @param my the y component of the predicted motion vector | |
825 */ | |
826 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){ | |
827 if(n==0){ | |
828 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ]; | |
829 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ]; | |
830 | |
1187 | 831 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list); |
1168 | 832 |
833 if(top_ref == ref){ | |
834 *mx= B[0]; | |
835 *my= B[1]; | |
836 return; | |
837 } | |
838 }else{ | |
839 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ]; | |
840 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ]; | |
841 | |
1187 | 842 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list); |
1168 | 843 |
844 if(left_ref == ref){ | |
845 *mx= A[0]; | |
846 *my= A[1]; | |
847 return; | |
848 } | |
849 } | |
850 | |
851 //RARE | |
852 pred_motion(h, n, 4, list, ref, mx, my); | |
853 } | |
854 | |
855 /** | |
856 * gets the directionally predicted 8x16 MV. | |
857 * @param n the block index | |
858 * @param mx the x component of the predicted motion vector | |
859 * @param my the y component of the predicted motion vector | |
860 */ | |
861 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){ | |
862 if(n==0){ | |
863 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ]; | |
864 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ]; | |
865 | |
1187 | 866 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list); |
1168 | 867 |
868 if(left_ref == ref){ | |
869 *mx= A[0]; | |
870 *my= A[1]; | |
871 return; | |
872 } | |
873 }else{ | |
1169 | 874 const int16_t * C; |
875 int diagonal_ref; | |
876 | |
877 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2); | |
1168 | 878 |
1187 | 879 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list); |
1168 | 880 |
1169 | 881 if(diagonal_ref == ref){ |
1168 | 882 *mx= C[0]; |
883 *my= C[1]; | |
884 return; | |
885 } | |
886 } | |
887 | |
888 //RARE | |
889 pred_motion(h, n, 2, list, ref, mx, my); | |
890 } | |
891 | |
892 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){ | |
893 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ]; | |
894 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ]; | |
895 | |
1187 | 896 tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y); |
1168 | 897 |
898 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE | |
899 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0) | |
900 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){ | |
901 | |
902 *mx = *my = 0; | |
903 return; | |
904 } | |
905 | |
906 pred_motion(h, 0, 4, 0, 0, mx, my); | |
907 | |
908 return; | |
909 } | |
910 | |
911 static inline void write_back_motion(H264Context *h, int mb_type){ | |
912 MpegEncContext * const s = &h->s; | |
913 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; | |
914 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride; | |
915 int list; | |
916 | |
917 for(list=0; list<2; list++){ | |
918 int y; | |
919 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){ | |
920 if(1){ //FIXME skip or never read if mb_type doesnt use it | |
921 for(y=0; y<4; y++){ | |
922 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= | |
923 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0; | |
924 } | |
925 for(y=0; y<2; y++){ | |
926 *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101; | |
927 } | |
928 } | |
929 continue; //FIXME direct mode ... | |
930 } | |
931 | |
932 for(y=0; y<4; y++){ | |
933 *(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]; | |
934 *(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]; | |
935 } | |
936 for(y=0; y<2; y++){ | |
937 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y]; | |
938 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y]; | |
939 } | |
940 } | |
941 } | |
942 | |
943 /** | |
944 * Decodes a network abstraction layer unit. | |
945 * @param consumed is the number of bytes used as input | |
946 * @param length is the length of the array | |
947 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing? | |
948 * @returns decoded bytes, might be src+1 if no escapes | |
949 */ | |
950 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){ | |
951 int i, si, di; | |
952 uint8_t *dst; | |
953 | |
954 // src[0]&0x80; //forbidden bit | |
955 h->nal_ref_idc= src[0]>>5; | |
956 h->nal_unit_type= src[0]&0x1F; | |
957 | |
958 src++; length--; | |
959 #if 0 | |
960 for(i=0; i<length; i++) | |
961 printf("%2X ", src[i]); | |
962 #endif | |
963 for(i=0; i+1<length; i+=2){ | |
964 if(src[i]) continue; | |
965 if(i>0 && src[i-1]==0) i--; | |
966 if(i+2<length && src[i+1]==0 && src[i+2]<=3){ | |
967 if(src[i+2]!=3){ | |
968 /* startcode, so we must be past the end */ | |
969 length=i; | |
970 } | |
971 break; | |
972 } | |
973 } | |
974 | |
975 if(i>=length-1){ //no escaped 0 | |
976 *dst_length= length; | |
977 *consumed= length+1; //+1 for the header | |
978 return src; | |
979 } | |
980 | |
981 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length); | |
982 dst= h->rbsp_buffer; | |
983 | |
984 //printf("deoding esc\n"); | |
985 si=di=0; | |
986 while(si<length){ | |
987 //remove escapes (very rare 1:2^22) | |
988 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){ | |
989 if(src[si+2]==3){ //escape | |
990 dst[di++]= 0; | |
991 dst[di++]= 0; | |
992 si+=3; | |
993 }else //next start code | |
994 break; | |
995 } | |
996 | |
997 dst[di++]= src[si++]; | |
998 } | |
999 | |
1000 *dst_length= di; | |
1001 *consumed= si + 1;//+1 for the header | |
1002 //FIXME store exact number of bits in the getbitcontext (its needed for decoding) | |
1003 return dst; | |
1004 } | |
1005 | |
1006 /** | |
1007 * @param src the data which should be escaped | |
1008 * @param dst the target buffer, dst+1 == src is allowed as a special case | |
1009 * @param length the length of the src data | |
1010 * @param dst_length the length of the dst array | |
1011 * @returns length of escaped data in bytes or -1 if an error occured | |
1012 */ | |
1013 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){ | |
1014 int i, escape_count, si, di; | |
1015 uint8_t *temp; | |
1016 | |
1017 assert(length>=0); | |
1018 assert(dst_length>0); | |
1019 | |
1020 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type; | |
1021 | |
1022 if(length==0) return 1; | |
1023 | |
1024 escape_count= 0; | |
1025 for(i=0; i<length; i+=2){ | |
1026 if(src[i]) continue; | |
1027 if(i>0 && src[i-1]==0) | |
1028 i--; | |
1029 if(i+2<length && src[i+1]==0 && src[i+2]<=3){ | |
1030 escape_count++; | |
1031 i+=2; | |
1032 } | |
1033 } | |
1034 | |
1035 if(escape_count==0){ | |
1036 if(dst+1 != src) | |
1037 memcpy(dst+1, src, length); | |
1038 return length + 1; | |
1039 } | |
1040 | |
1041 if(length + escape_count + 1> dst_length) | |
1042 return -1; | |
1043 | |
1044 //this should be damn rare (hopefully) | |
1045 | |
1046 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count); | |
1047 temp= h->rbsp_buffer; | |
1048 //printf("encoding esc\n"); | |
1049 | |
1050 si= 0; | |
1051 di= 0; | |
1052 while(si < length){ | |
1053 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){ | |
1054 temp[di++]= 0; si++; | |
1055 temp[di++]= 0; si++; | |
1056 temp[di++]= 3; | |
1057 temp[di++]= src[si++]; | |
1058 } | |
1059 else | |
1060 temp[di++]= src[si++]; | |
1061 } | |
1062 memcpy(dst+1, temp, length+escape_count); | |
1063 | |
1064 assert(di == length+escape_count); | |
1065 | |
1066 return di + 1; | |
1067 } | |
1068 | |
1069 /** | |
1070 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4 | |
1071 */ | |
1072 static void encode_rbsp_trailing(PutBitContext *pb){ | |
1073 int length; | |
1074 put_bits(pb, 1, 1); | |
1075 length= (-get_bit_count(pb))&7; | |
1076 if(length) put_bits(pb, length, 0); | |
1077 } | |
1078 | |
1079 /** | |
1080 * identifies the exact end of the bitstream | |
1081 * @return the length of the trailing, or 0 if damaged | |
1082 */ | |
1083 static int decode_rbsp_trailing(uint8_t *src){ | |
1084 int v= *src; | |
1085 int r; | |
1086 | |
1170 | 1087 tprintf("rbsp trailing %X\n", v); |
1168 | 1088 |
1089 for(r=1; r<9; r++){ | |
1090 if(v&1) return r; | |
1091 v>>=1; | |
1092 } | |
1093 return 0; | |
1094 } | |
1095 | |
1096 /** | |
1097 * idct tranforms the 16 dc values and dequantize them. | |
1098 * @param qp quantization parameter | |
1099 */ | |
1100 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){ | |
1101 const int qmul= dequant_coeff[qp][0]; | |
1102 #define stride 16 | |
1103 int i; | |
1104 int temp[16]; //FIXME check if this is a good idea | |
1105 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride}; | |
1106 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride}; | |
1107 | |
1108 //memset(block, 64, 2*256); | |
1109 //return; | |
1110 for(i=0; i<4; i++){ | |
1111 const int offset= y_offset[i]; | |
1112 const int z0= block[offset+stride*0] + block[offset+stride*4]; | |
1113 const int z1= block[offset+stride*0] - block[offset+stride*4]; | |
1114 const int z2= block[offset+stride*1] - block[offset+stride*5]; | |
1115 const int z3= block[offset+stride*1] + block[offset+stride*5]; | |
1116 | |
1117 temp[4*i+0]= z0+z3; | |
1118 temp[4*i+1]= z1+z2; | |
1119 temp[4*i+2]= z1-z2; | |
1120 temp[4*i+3]= z0-z3; | |
1121 } | |
1122 | |
1123 for(i=0; i<4; i++){ | |
1124 const int offset= x_offset[i]; | |
1125 const int z0= temp[4*0+i] + temp[4*2+i]; | |
1126 const int z1= temp[4*0+i] - temp[4*2+i]; | |
1127 const int z2= temp[4*1+i] - temp[4*3+i]; | |
1128 const int z3= temp[4*1+i] + temp[4*3+i]; | |
1129 | |
1130 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual | |
1131 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2; | |
1132 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2; | |
1133 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2; | |
1134 } | |
1135 } | |
1136 | |
1137 /** | |
1138 * dct tranforms the 16 dc values. | |
1139 * @param qp quantization parameter ??? FIXME | |
1140 */ | |
1141 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){ | |
1142 // const int qmul= dequant_coeff[qp][0]; | |
1143 int i; | |
1144 int temp[16]; //FIXME check if this is a good idea | |
1145 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride}; | |
1146 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride}; | |
1147 | |
1148 for(i=0; i<4; i++){ | |
1149 const int offset= y_offset[i]; | |
1150 const int z0= block[offset+stride*0] + block[offset+stride*4]; | |
1151 const int z1= block[offset+stride*0] - block[offset+stride*4]; | |
1152 const int z2= block[offset+stride*1] - block[offset+stride*5]; | |
1153 const int z3= block[offset+stride*1] + block[offset+stride*5]; | |
1154 | |
1155 temp[4*i+0]= z0+z3; | |
1156 temp[4*i+1]= z1+z2; | |
1157 temp[4*i+2]= z1-z2; | |
1158 temp[4*i+3]= z0-z3; | |
1159 } | |
1160 | |
1161 for(i=0; i<4; i++){ | |
1162 const int offset= x_offset[i]; | |
1163 const int z0= temp[4*0+i] + temp[4*2+i]; | |
1164 const int z1= temp[4*0+i] - temp[4*2+i]; | |
1165 const int z2= temp[4*1+i] - temp[4*3+i]; | |
1166 const int z3= temp[4*1+i] + temp[4*3+i]; | |
1167 | |
1168 block[stride*0 +offset]= (z0 + z3)>>1; | |
1169 block[stride*2 +offset]= (z1 + z2)>>1; | |
1170 block[stride*8 +offset]= (z1 - z2)>>1; | |
1171 block[stride*10+offset]= (z0 - z3)>>1; | |
1172 } | |
1173 } | |
1174 #undef xStride | |
1175 #undef stride | |
1176 | |
1177 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){ | |
1178 const int qmul= dequant_coeff[qp][0]; | |
1179 const int stride= 16*2; | |
1180 const int xStride= 16; | |
1181 int a,b,c,d,e; | |
1182 | |
1183 a= block[stride*0 + xStride*0]; | |
1184 b= block[stride*0 + xStride*1]; | |
1185 c= block[stride*1 + xStride*0]; | |
1186 d= block[stride*1 + xStride*1]; | |
1187 | |
1188 e= a-b; | |
1189 a= a+b; | |
1190 b= c-d; | |
1191 c= c+d; | |
1192 | |
1193 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1; | |
1194 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1; | |
1195 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1; | |
1196 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1; | |
1197 } | |
1198 | |
1199 static void chroma_dc_dct_c(DCTELEM *block){ | |
1200 const int stride= 16*2; | |
1201 const int xStride= 16; | |
1202 int a,b,c,d,e; | |
1203 | |
1204 a= block[stride*0 + xStride*0]; | |
1205 b= block[stride*0 + xStride*1]; | |
1206 c= block[stride*1 + xStride*0]; | |
1207 d= block[stride*1 + xStride*1]; | |
1208 | |
1209 e= a-b; | |
1210 a= a+b; | |
1211 b= c-d; | |
1212 c= c+d; | |
1213 | |
1214 block[stride*0 + xStride*0]= (a+c); | |
1215 block[stride*0 + xStride*1]= (e+b); | |
1216 block[stride*1 + xStride*0]= (a-c); | |
1217 block[stride*1 + xStride*1]= (e-b); | |
1218 } | |
1219 | |
1220 /** | |
1221 * gets the chroma qp. | |
1222 */ | |
1223 static inline int get_chroma_qp(H264Context *h, int qscale){ | |
1224 | |
1225 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)]; | |
1226 } | |
1227 | |
1228 | |
1229 /** | |
1230 * | |
1231 */ | |
1232 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){ | |
1233 int i; | |
1234 uint8_t *cm = cropTbl + MAX_NEG_CROP; | |
1235 | |
1236 block[0] += 32; | |
1237 #if 1 | |
1238 for(i=0; i<4; i++){ | |
1239 const int z0= block[i + 4*0] + block[i + 4*2]; | |
1240 const int z1= block[i + 4*0] - block[i + 4*2]; | |
1241 const int z2= (block[i + 4*1]>>1) - block[i + 4*3]; | |
1242 const int z3= block[i + 4*1] + (block[i + 4*3]>>1); | |
1243 | |
1244 block[i + 4*0]= z0 + z3; | |
1245 block[i + 4*1]= z1 + z2; | |
1246 block[i + 4*2]= z1 - z2; | |
1247 block[i + 4*3]= z0 - z3; | |
1248 } | |
1249 | |
1250 for(i=0; i<4; i++){ | |
1251 const int z0= block[0 + 4*i] + block[2 + 4*i]; | |
1252 const int z1= block[0 + 4*i] - block[2 + 4*i]; | |
1253 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i]; | |
1254 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1); | |
1255 | |
1256 dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ]; | |
1257 dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ]; | |
1258 dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ]; | |
1259 dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ]; | |
1260 } | |
1261 #else | |
1262 for(i=0; i<4; i++){ | |
1263 const int z0= block[0 + 4*i] + block[2 + 4*i]; | |
1264 const int z1= block[0 + 4*i] - block[2 + 4*i]; | |
1265 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i]; | |
1266 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1); | |
1267 | |
1268 block[0 + 4*i]= z0 + z3; | |
1269 block[1 + 4*i]= z1 + z2; | |
1270 block[2 + 4*i]= z1 - z2; | |
1271 block[3 + 4*i]= z0 - z3; | |
1272 } | |
1273 | |
1274 for(i=0; i<4; i++){ | |
1275 const int z0= block[i + 4*0] + block[i + 4*2]; | |
1276 const int z1= block[i + 4*0] - block[i + 4*2]; | |
1277 const int z2= (block[i + 4*1]>>1) - block[i + 4*3]; | |
1278 const int z3= block[i + 4*1] + (block[i + 4*3]>>1); | |
1279 | |
1280 dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ]; | |
1281 dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ]; | |
1282 dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ]; | |
1283 dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ]; | |
1284 } | |
1285 #endif | |
1286 } | |
1287 | |
1288 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){ | |
1289 int i; | |
1290 //FIXME try int temp instead of block | |
1291 | |
1292 for(i=0; i<4; i++){ | |
1293 const int d0= src1[0 + i*stride] - src2[0 + i*stride]; | |
1294 const int d1= src1[1 + i*stride] - src2[1 + i*stride]; | |
1295 const int d2= src1[2 + i*stride] - src2[2 + i*stride]; | |
1296 const int d3= src1[3 + i*stride] - src2[3 + i*stride]; | |
1297 const int z0= d0 + d3; | |
1298 const int z3= d0 - d3; | |
1299 const int z1= d1 + d2; | |
1300 const int z2= d1 - d2; | |
1301 | |
1302 block[0 + 4*i]= z0 + z1; | |
1303 block[1 + 4*i]= 2*z3 + z2; | |
1304 block[2 + 4*i]= z0 - z1; | |
1305 block[3 + 4*i]= z3 - 2*z2; | |
1306 } | |
1307 | |
1308 for(i=0; i<4; i++){ | |
1309 const int z0= block[0*4 + i] + block[3*4 + i]; | |
1310 const int z3= block[0*4 + i] - block[3*4 + i]; | |
1311 const int z1= block[1*4 + i] + block[2*4 + i]; | |
1312 const int z2= block[1*4 + i] - block[2*4 + i]; | |
1313 | |
1314 block[0*4 + i]= z0 + z1; | |
1315 block[1*4 + i]= 2*z3 + z2; | |
1316 block[2*4 + i]= z0 - z1; | |
1317 block[3*4 + i]= z3 - 2*z2; | |
1318 } | |
1319 } | |
1320 | |
1321 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close | |
1322 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away) | |
1323 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){ | |
1324 int i; | |
1325 const int * const quant_table= quant_coeff[qscale]; | |
1326 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6; | |
1327 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1; | |
1328 const unsigned int threshold2= (threshold1<<1); | |
1329 int last_non_zero; | |
1330 | |
1331 if(seperate_dc){ | |
1332 if(qscale<=18){ | |
1333 //avoid overflows | |
1334 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6; | |
1335 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1; | |
1336 const unsigned int dc_threshold2= (dc_threshold1<<1); | |
1337 | |
1338 int level= block[0]*quant_coeff[qscale+18][0]; | |
1339 if(((unsigned)(level+dc_threshold1))>dc_threshold2){ | |
1340 if(level>0){ | |
1341 level= (dc_bias + level)>>(QUANT_SHIFT-2); | |
1342 block[0]= level; | |
1343 }else{ | |
1344 level= (dc_bias - level)>>(QUANT_SHIFT-2); | |
1345 block[0]= -level; | |
1346 } | |
1347 // last_non_zero = i; | |
1348 }else{ | |
1349 block[0]=0; | |
1350 } | |
1351 }else{ | |
1352 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6; | |
1353 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1; | |
1354 const unsigned int dc_threshold2= (dc_threshold1<<1); | |
1355 | |
1356 int level= block[0]*quant_table[0]; | |
1357 if(((unsigned)(level+dc_threshold1))>dc_threshold2){ | |
1358 if(level>0){ | |
1359 level= (dc_bias + level)>>(QUANT_SHIFT+1); | |
1360 block[0]= level; | |
1361 }else{ | |
1362 level= (dc_bias - level)>>(QUANT_SHIFT+1); | |
1363 block[0]= -level; | |
1364 } | |
1365 // last_non_zero = i; | |
1366 }else{ | |
1367 block[0]=0; | |
1368 } | |
1369 } | |
1370 last_non_zero= 0; | |
1371 i=1; | |
1372 }else{ | |
1373 last_non_zero= -1; | |
1374 i=0; | |
1375 } | |
1376 | |
1377 for(; i<16; i++){ | |
1378 const int j= scantable[i]; | |
1379 int level= block[j]*quant_table[j]; | |
1380 | |
1381 // if( bias+level >= (1<<(QMAT_SHIFT - 3)) | |
1382 // || bias-level >= (1<<(QMAT_SHIFT - 3))){ | |
1383 if(((unsigned)(level+threshold1))>threshold2){ | |
1384 if(level>0){ | |
1385 level= (bias + level)>>QUANT_SHIFT; | |
1386 block[j]= level; | |
1387 }else{ | |
1388 level= (bias - level)>>QUANT_SHIFT; | |
1389 block[j]= -level; | |
1390 } | |
1391 last_non_zero = i; | |
1392 }else{ | |
1393 block[j]=0; | |
1394 } | |
1395 } | |
1396 | |
1397 return last_non_zero; | |
1398 } | |
1399 | |
1400 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){ | |
1401 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
1402 ((uint32_t*)(src+0*stride))[0]= a; | |
1403 ((uint32_t*)(src+1*stride))[0]= a; | |
1404 ((uint32_t*)(src+2*stride))[0]= a; | |
1405 ((uint32_t*)(src+3*stride))[0]= a; | |
1406 } | |
1407 | |
1408 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){ | |
1409 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101; | |
1410 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101; | |
1411 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101; | |
1412 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101; | |
1413 } | |
1414 | |
1415 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1416 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] | |
1417 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3; | |
1418 | |
1419 ((uint32_t*)(src+0*stride))[0]= | |
1420 ((uint32_t*)(src+1*stride))[0]= | |
1421 ((uint32_t*)(src+2*stride))[0]= | |
1422 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1423 } | |
1424 | |
1425 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1426 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2; | |
1427 | |
1428 ((uint32_t*)(src+0*stride))[0]= | |
1429 ((uint32_t*)(src+1*stride))[0]= | |
1430 ((uint32_t*)(src+2*stride))[0]= | |
1431 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1432 } | |
1433 | |
1434 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1435 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2; | |
1436 | |
1437 ((uint32_t*)(src+0*stride))[0]= | |
1438 ((uint32_t*)(src+1*stride))[0]= | |
1439 ((uint32_t*)(src+2*stride))[0]= | |
1440 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1441 } | |
1442 | |
1443 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1444 ((uint32_t*)(src+0*stride))[0]= | |
1445 ((uint32_t*)(src+1*stride))[0]= | |
1446 ((uint32_t*)(src+2*stride))[0]= | |
1447 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U; | |
1448 } | |
1449 | |
1450 | |
1451 #define LOAD_TOP_RIGHT_EDGE\ | |
1452 const int t4= topright[0];\ | |
1453 const int t5= topright[1];\ | |
1454 const int t6= topright[2];\ | |
1455 const int t7= topright[3];\ | |
1456 | |
1457 #define LOAD_LEFT_EDGE\ | |
1458 const int l0= src[-1+0*stride];\ | |
1459 const int l1= src[-1+1*stride];\ | |
1460 const int l2= src[-1+2*stride];\ | |
1461 const int l3= src[-1+3*stride];\ | |
1462 | |
1463 #define LOAD_TOP_EDGE\ | |
1464 const int t0= src[ 0-1*stride];\ | |
1465 const int t1= src[ 1-1*stride];\ | |
1466 const int t2= src[ 2-1*stride];\ | |
1467 const int t3= src[ 3-1*stride];\ | |
1468 | |
1469 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){ | |
1470 const int lt= src[-1-1*stride]; | |
1471 LOAD_TOP_EDGE | |
1472 LOAD_LEFT_EDGE | |
1473 | |
1474 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; | |
1475 src[0+2*stride]= | |
1476 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; | |
1477 src[0+1*stride]= | |
1478 src[1+2*stride]= | |
1479 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; | |
1480 src[0+0*stride]= | |
1481 src[1+1*stride]= | |
1482 src[2+2*stride]= | |
1483 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
1484 src[1+0*stride]= | |
1485 src[2+1*stride]= | |
1486 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
1487 src[2+0*stride]= | |
1488 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1489 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1282 | 1490 } |
1168 | 1491 |
1492 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){ | |
1493 LOAD_TOP_EDGE | |
1494 LOAD_TOP_RIGHT_EDGE | |
1495 // LOAD_LEFT_EDGE | |
1496 | |
1497 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2; | |
1498 src[1+0*stride]= | |
1499 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2; | |
1500 src[2+0*stride]= | |
1501 src[1+1*stride]= | |
1502 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2; | |
1503 src[3+0*stride]= | |
1504 src[2+1*stride]= | |
1505 src[1+2*stride]= | |
1506 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2; | |
1507 src[3+1*stride]= | |
1508 src[2+2*stride]= | |
1509 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2; | |
1510 src[3+2*stride]= | |
1511 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2; | |
1512 src[3+3*stride]=(t6 + 3*t7 + 2)>>2; | |
1282 | 1513 } |
1168 | 1514 |
1515 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){ | |
1516 const int lt= src[-1-1*stride]; | |
1517 LOAD_TOP_EDGE | |
1518 LOAD_LEFT_EDGE | |
1519 const __attribute__((unused)) int unu= l3; | |
1520 | |
1521 src[0+0*stride]= | |
1522 src[1+2*stride]=(lt + t0 + 1)>>1; | |
1523 src[1+0*stride]= | |
1524 src[2+2*stride]=(t0 + t1 + 1)>>1; | |
1525 src[2+0*stride]= | |
1526 src[3+2*stride]=(t1 + t2 + 1)>>1; | |
1527 src[3+0*stride]=(t2 + t3 + 1)>>1; | |
1528 src[0+1*stride]= | |
1529 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
1530 src[1+1*stride]= | |
1531 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
1532 src[2+1*stride]= | |
1533 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1534 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1535 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2; | |
1536 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1282 | 1537 } |
1168 | 1538 |
1539 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){ | |
1540 LOAD_TOP_EDGE | |
1541 LOAD_TOP_RIGHT_EDGE | |
1542 const __attribute__((unused)) int unu= t7; | |
1543 | |
1544 src[0+0*stride]=(t0 + t1 + 1)>>1; | |
1545 src[1+0*stride]= | |
1546 src[0+2*stride]=(t1 + t2 + 1)>>1; | |
1547 src[2+0*stride]= | |
1548 src[1+2*stride]=(t2 + t3 + 1)>>1; | |
1549 src[3+0*stride]= | |
1550 src[2+2*stride]=(t3 + t4+ 1)>>1; | |
1551 src[3+2*stride]=(t4 + t5+ 1)>>1; | |
1552 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1553 src[1+1*stride]= | |
1554 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1555 src[2+1*stride]= | |
1556 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2; | |
1557 src[3+1*stride]= | |
1558 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2; | |
1559 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2; | |
1282 | 1560 } |
1168 | 1561 |
1562 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){ | |
1563 LOAD_LEFT_EDGE | |
1564 | |
1565 src[0+0*stride]=(l0 + l1 + 1)>>1; | |
1566 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1567 src[2+0*stride]= | |
1568 src[0+1*stride]=(l1 + l2 + 1)>>1; | |
1569 src[3+0*stride]= | |
1570 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2; | |
1571 src[2+1*stride]= | |
1572 src[0+2*stride]=(l2 + l3 + 1)>>1; | |
1573 src[3+1*stride]= | |
1574 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2; | |
1575 src[3+2*stride]= | |
1576 src[1+3*stride]= | |
1577 src[0+3*stride]= | |
1578 src[2+2*stride]= | |
1579 src[2+3*stride]= | |
1580 src[3+3*stride]=l3; | |
1282 | 1581 } |
1168 | 1582 |
1583 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){ | |
1584 const int lt= src[-1-1*stride]; | |
1585 LOAD_TOP_EDGE | |
1586 LOAD_LEFT_EDGE | |
1587 const __attribute__((unused)) int unu= t3; | |
1588 | |
1589 src[0+0*stride]= | |
1590 src[2+1*stride]=(lt + l0 + 1)>>1; | |
1591 src[1+0*stride]= | |
1592 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
1593 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
1594 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1595 src[0+1*stride]= | |
1596 src[2+2*stride]=(l0 + l1 + 1)>>1; | |
1597 src[1+1*stride]= | |
1598 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2; | |
1599 src[0+2*stride]= | |
1600 src[2+3*stride]=(l1 + l2+ 1)>>1; | |
1601 src[1+2*stride]= | |
1602 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1603 src[0+3*stride]=(l2 + l3 + 1)>>1; | |
1604 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2; | |
1282 | 1605 } |
1168 | 1606 |
1607 static void pred16x16_vertical_c(uint8_t *src, int stride){ | |
1608 int i; | |
1609 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
1610 const uint32_t b= ((uint32_t*)(src-stride))[1]; | |
1611 const uint32_t c= ((uint32_t*)(src-stride))[2]; | |
1612 const uint32_t d= ((uint32_t*)(src-stride))[3]; | |
1613 | |
1614 for(i=0; i<16; i++){ | |
1615 ((uint32_t*)(src+i*stride))[0]= a; | |
1616 ((uint32_t*)(src+i*stride))[1]= b; | |
1617 ((uint32_t*)(src+i*stride))[2]= c; | |
1618 ((uint32_t*)(src+i*stride))[3]= d; | |
1619 } | |
1620 } | |
1621 | |
1622 static void pred16x16_horizontal_c(uint8_t *src, int stride){ | |
1623 int i; | |
1624 | |
1625 for(i=0; i<16; i++){ | |
1626 ((uint32_t*)(src+i*stride))[0]= | |
1627 ((uint32_t*)(src+i*stride))[1]= | |
1628 ((uint32_t*)(src+i*stride))[2]= | |
1629 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101; | |
1630 } | |
1631 } | |
1632 | |
1633 static void pred16x16_dc_c(uint8_t *src, int stride){ | |
1634 int i, dc=0; | |
1635 | |
1636 for(i=0;i<16; i++){ | |
1637 dc+= src[-1+i*stride]; | |
1638 } | |
1639 | |
1640 for(i=0;i<16; i++){ | |
1641 dc+= src[i-stride]; | |
1642 } | |
1643 | |
1644 dc= 0x01010101*((dc + 16)>>5); | |
1645 | |
1646 for(i=0; i<16; i++){ | |
1647 ((uint32_t*)(src+i*stride))[0]= | |
1648 ((uint32_t*)(src+i*stride))[1]= | |
1649 ((uint32_t*)(src+i*stride))[2]= | |
1650 ((uint32_t*)(src+i*stride))[3]= dc; | |
1651 } | |
1652 } | |
1653 | |
1654 static void pred16x16_left_dc_c(uint8_t *src, int stride){ | |
1655 int i, dc=0; | |
1656 | |
1657 for(i=0;i<16; i++){ | |
1658 dc+= src[-1+i*stride]; | |
1659 } | |
1660 | |
1661 dc= 0x01010101*((dc + 8)>>4); | |
1662 | |
1663 for(i=0; i<16; i++){ | |
1664 ((uint32_t*)(src+i*stride))[0]= | |
1665 ((uint32_t*)(src+i*stride))[1]= | |
1666 ((uint32_t*)(src+i*stride))[2]= | |
1667 ((uint32_t*)(src+i*stride))[3]= dc; | |
1668 } | |
1669 } | |
1670 | |
1671 static void pred16x16_top_dc_c(uint8_t *src, int stride){ | |
1672 int i, dc=0; | |
1673 | |
1674 for(i=0;i<16; i++){ | |
1675 dc+= src[i-stride]; | |
1676 } | |
1677 dc= 0x01010101*((dc + 8)>>4); | |
1678 | |
1679 for(i=0; i<16; i++){ | |
1680 ((uint32_t*)(src+i*stride))[0]= | |
1681 ((uint32_t*)(src+i*stride))[1]= | |
1682 ((uint32_t*)(src+i*stride))[2]= | |
1683 ((uint32_t*)(src+i*stride))[3]= dc; | |
1684 } | |
1685 } | |
1686 | |
1687 static void pred16x16_128_dc_c(uint8_t *src, int stride){ | |
1688 int i; | |
1689 | |
1690 for(i=0; i<16; i++){ | |
1691 ((uint32_t*)(src+i*stride))[0]= | |
1692 ((uint32_t*)(src+i*stride))[1]= | |
1693 ((uint32_t*)(src+i*stride))[2]= | |
1694 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U; | |
1695 } | |
1696 } | |
1697 | |
1234 | 1698 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){ |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1699 int i, j, k; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1700 int a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1701 uint8_t *cm = cropTbl + MAX_NEG_CROP; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1702 const uint8_t * const src0 = src+7-stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1703 const uint8_t *src1 = src+8*stride-1; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1704 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1705 int H = src0[1] - src0[-1]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1706 int V = src1[0] - src2[ 0]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1707 for(k=2; k<=8; ++k) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1708 src1 += stride; src2 -= stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1709 H += k*(src0[k] - src0[-k]); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1710 V += k*(src1[0] - src2[ 0]); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1711 } |
1234 | 1712 if(svq3){ |
1713 H = ( 5*(H/4) ) / 16; | |
1714 V = ( 5*(V/4) ) / 16; | |
1330
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
1715 |
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
1716 /* required for 100% accuracy */ |
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
1717 i = H; H = V; V = i; |
1234 | 1718 }else{ |
1719 H = ( 5*H+32 ) >> 6; | |
1720 V = ( 5*V+32 ) >> 6; | |
1721 } | |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1722 |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1723 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1724 for(j=16; j>0; --j) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1725 int b = a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1726 a += V; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1727 for(i=-16; i<0; i+=4) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1728 src[16+i] = cm[ (b ) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1729 src[17+i] = cm[ (b+ H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1730 src[18+i] = cm[ (b+2*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1731 src[19+i] = cm[ (b+3*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1732 b += 4*H; |
1168 | 1733 } |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1734 src += stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1735 } |
1168 | 1736 } |
1737 | |
1234 | 1738 static void pred16x16_plane_c(uint8_t *src, int stride){ |
1739 pred16x16_plane_compat_c(src, stride, 0); | |
1740 } | |
1741 | |
1168 | 1742 static void pred8x8_vertical_c(uint8_t *src, int stride){ |
1743 int i; | |
1744 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
1745 const uint32_t b= ((uint32_t*)(src-stride))[1]; | |
1746 | |
1747 for(i=0; i<8; i++){ | |
1748 ((uint32_t*)(src+i*stride))[0]= a; | |
1749 ((uint32_t*)(src+i*stride))[1]= b; | |
1750 } | |
1751 } | |
1752 | |
1753 static void pred8x8_horizontal_c(uint8_t *src, int stride){ | |
1754 int i; | |
1755 | |
1756 for(i=0; i<8; i++){ | |
1757 ((uint32_t*)(src+i*stride))[0]= | |
1758 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101; | |
1759 } | |
1760 } | |
1761 | |
1762 static void pred8x8_128_dc_c(uint8_t *src, int stride){ | |
1763 int i; | |
1764 | |
1765 for(i=0; i<4; i++){ | |
1766 ((uint32_t*)(src+i*stride))[0]= | |
1767 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U; | |
1768 } | |
1769 for(i=4; i<8; i++){ | |
1770 ((uint32_t*)(src+i*stride))[0]= | |
1771 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U; | |
1772 } | |
1773 } | |
1774 | |
1775 static void pred8x8_left_dc_c(uint8_t *src, int stride){ | |
1776 int i; | |
1777 int dc0, dc2; | |
1778 | |
1779 dc0=dc2=0; | |
1780 for(i=0;i<4; i++){ | |
1781 dc0+= src[-1+i*stride]; | |
1782 dc2+= src[-1+(i+4)*stride]; | |
1783 } | |
1784 dc0= 0x01010101*((dc0 + 2)>>2); | |
1785 dc2= 0x01010101*((dc2 + 2)>>2); | |
1786 | |
1787 for(i=0; i<4; i++){ | |
1788 ((uint32_t*)(src+i*stride))[0]= | |
1789 ((uint32_t*)(src+i*stride))[1]= dc0; | |
1790 } | |
1791 for(i=4; i<8; i++){ | |
1792 ((uint32_t*)(src+i*stride))[0]= | |
1793 ((uint32_t*)(src+i*stride))[1]= dc2; | |
1794 } | |
1795 } | |
1796 | |
1797 static void pred8x8_top_dc_c(uint8_t *src, int stride){ | |
1798 int i; | |
1799 int dc0, dc1; | |
1800 | |
1801 dc0=dc1=0; | |
1802 for(i=0;i<4; i++){ | |
1803 dc0+= src[i-stride]; | |
1804 dc1+= src[4+i-stride]; | |
1805 } | |
1806 dc0= 0x01010101*((dc0 + 2)>>2); | |
1807 dc1= 0x01010101*((dc1 + 2)>>2); | |
1808 | |
1809 for(i=0; i<4; i++){ | |
1810 ((uint32_t*)(src+i*stride))[0]= dc0; | |
1811 ((uint32_t*)(src+i*stride))[1]= dc1; | |
1812 } | |
1813 for(i=4; i<8; i++){ | |
1814 ((uint32_t*)(src+i*stride))[0]= dc0; | |
1815 ((uint32_t*)(src+i*stride))[1]= dc1; | |
1816 } | |
1817 } | |
1818 | |
1819 | |
1820 static void pred8x8_dc_c(uint8_t *src, int stride){ | |
1821 int i; | |
1822 int dc0, dc1, dc2, dc3; | |
1823 | |
1824 dc0=dc1=dc2=0; | |
1825 for(i=0;i<4; i++){ | |
1826 dc0+= src[-1+i*stride] + src[i-stride]; | |
1827 dc1+= src[4+i-stride]; | |
1828 dc2+= src[-1+(i+4)*stride]; | |
1829 } | |
1830 dc3= 0x01010101*((dc1 + dc2 + 4)>>3); | |
1831 dc0= 0x01010101*((dc0 + 4)>>3); | |
1832 dc1= 0x01010101*((dc1 + 2)>>2); | |
1833 dc2= 0x01010101*((dc2 + 2)>>2); | |
1834 | |
1835 for(i=0; i<4; i++){ | |
1836 ((uint32_t*)(src+i*stride))[0]= dc0; | |
1837 ((uint32_t*)(src+i*stride))[1]= dc1; | |
1838 } | |
1839 for(i=4; i<8; i++){ | |
1840 ((uint32_t*)(src+i*stride))[0]= dc2; | |
1841 ((uint32_t*)(src+i*stride))[1]= dc3; | |
1842 } | |
1843 } | |
1844 | |
1845 static void pred8x8_plane_c(uint8_t *src, int stride){ | |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1846 int j, k; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1847 int a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1848 uint8_t *cm = cropTbl + MAX_NEG_CROP; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1849 const uint8_t * const src0 = src+3-stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1850 const uint8_t *src1 = src+4*stride-1; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1851 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1852 int H = src0[1] - src0[-1]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1853 int V = src1[0] - src2[ 0]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1854 for(k=2; k<=4; ++k) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1855 src1 += stride; src2 -= stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1856 H += k*(src0[k] - src0[-k]); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1857 V += k*(src1[0] - src2[ 0]); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1858 } |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1859 H = ( 17*H+16 ) >> 5; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1860 V = ( 17*V+16 ) >> 5; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1861 |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1862 a = 16*(src1[0] + src2[8]+1) - 3*(V+H); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1863 for(j=8; j>0; --j) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1864 int b = a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1865 a += V; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1866 src[0] = cm[ (b ) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1867 src[1] = cm[ (b+ H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1868 src[2] = cm[ (b+2*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1869 src[3] = cm[ (b+3*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1870 src[4] = cm[ (b+4*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1871 src[5] = cm[ (b+5*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1872 src[6] = cm[ (b+6*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1873 src[7] = cm[ (b+7*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1874 src += stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1875 } |
1168 | 1876 } |
1877 | |
1878 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list, | |
1879 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
1880 int src_x_offset, int src_y_offset, | |
1881 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){ | |
1882 MpegEncContext * const s = &h->s; | |
1883 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8; | |
1884 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8; | |
1885 const int luma_xy= (mx&3) + ((my&3)<<2); | |
1886 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize; | |
1887 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize; | |
1888 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize; | |
1889 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it | |
1890 int extra_height= extra_width; | |
1891 int emu=0; | |
1892 const int full_mx= mx>>2; | |
1893 const int full_my= my>>2; | |
1894 | |
1895 assert(pic->data[0]); | |
1896 | |
1897 if(mx&7) extra_width -= 3; | |
1898 if(my&7) extra_height -= 3; | |
1899 | |
1900 if( full_mx < 0-extra_width | |
1901 || full_my < 0-extra_height | |
1902 || full_mx + 16/*FIXME*/ > s->width + extra_width | |
1903 || full_my + 16/*FIXME*/ > s->height + extra_height){ | |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1282
diff
changeset
|
1904 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height); |
1168 | 1905 src_y= s->edge_emu_buffer + 2 + 2*s->linesize; |
1906 emu=1; | |
1907 } | |
1908 | |
1909 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps? | |
1910 if(!square){ | |
1911 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize); | |
1912 } | |
1913 | |
1914 if(s->flags&CODEC_FLAG_GRAY) return; | |
1915 | |
1916 if(emu){ | |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1282
diff
changeset
|
1917 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1); |
1168 | 1918 src_cb= s->edge_emu_buffer; |
1919 } | |
1920 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7); | |
1921 | |
1922 if(emu){ | |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1282
diff
changeset
|
1923 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1); |
1168 | 1924 src_cr= s->edge_emu_buffer; |
1925 } | |
1926 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7); | |
1927 } | |
1928 | |
1929 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta, | |
1930 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
1931 int x_offset, int y_offset, | |
1932 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put, | |
1933 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg, | |
1934 int list0, int list1){ | |
1935 MpegEncContext * const s = &h->s; | |
1936 qpel_mc_func *qpix_op= qpix_put; | |
1937 h264_chroma_mc_func chroma_op= chroma_put; | |
1938 | |
1939 dest_y += 2*x_offset + 2*y_offset*s-> linesize; | |
1940 dest_cb += x_offset + y_offset*s->uvlinesize; | |
1941 dest_cr += x_offset + y_offset*s->uvlinesize; | |
1942 x_offset += 8*s->mb_x; | |
1943 y_offset += 8*s->mb_y; | |
1944 | |
1945 if(list0){ | |
1169 | 1946 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ]; |
1168 | 1947 mc_dir_part(h, ref, n, square, chroma_height, delta, 0, |
1948 dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
1949 qpix_op, chroma_op); | |
1950 | |
1951 qpix_op= qpix_avg; | |
1952 chroma_op= chroma_avg; | |
1953 } | |
1954 | |
1955 if(list1){ | |
1169 | 1956 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ]; |
1168 | 1957 mc_dir_part(h, ref, n, square, chroma_height, delta, 1, |
1958 dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
1959 qpix_op, chroma_op); | |
1960 } | |
1961 } | |
1962 | |
1963 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
1964 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put), | |
1965 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){ | |
1966 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
1967 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 1968 const int mb_type= s->current_picture.mb_type[mb_xy]; |
1969 | |
1970 assert(IS_INTER(mb_type)); | |
1971 | |
1972 if(IS_16X16(mb_type)){ | |
1973 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0, | |
1974 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0], | |
1975 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); | |
1976 }else if(IS_16X8(mb_type)){ | |
1977 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0, | |
1978 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0], | |
1979 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); | |
1980 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4, | |
1981 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0], | |
1982 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1)); | |
1983 }else if(IS_8X16(mb_type)){ | |
1984 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0, | |
1985 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], | |
1986 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); | |
1987 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0, | |
1988 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], | |
1989 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1)); | |
1990 }else{ | |
1991 int i; | |
1992 | |
1993 assert(IS_8X8(mb_type)); | |
1994 | |
1995 for(i=0; i<4; i++){ | |
1996 const int sub_mb_type= h->sub_mb_type[i]; | |
1997 const int n= 4*i; | |
1998 int x_offset= (i&1)<<2; | |
1999 int y_offset= (i&2)<<1; | |
2000 | |
2001 if(IS_SUB_8X8(sub_mb_type)){ | |
2002 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2003 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], | |
2004 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2005 }else if(IS_SUB_8X4(sub_mb_type)){ | |
2006 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2007 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1], | |
2008 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2009 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2, | |
2010 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1], | |
2011 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2012 }else if(IS_SUB_4X8(sub_mb_type)){ | |
2013 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2014 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], | |
2015 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2016 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset, | |
2017 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], | |
2018 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2019 }else{ | |
2020 int j; | |
2021 assert(IS_SUB_4X4(sub_mb_type)); | |
2022 for(j=0; j<4; j++){ | |
2023 int sub_x_offset= x_offset + 2*(j&1); | |
2024 int sub_y_offset= y_offset + (j&2); | |
2025 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset, | |
2026 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], | |
2027 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2028 } | |
2029 } | |
2030 } | |
2031 } | |
2032 } | |
2033 | |
2034 static void decode_init_vlc(H264Context *h){ | |
2035 static int done = 0; | |
2036 | |
2037 if (!done) { | |
2038 int i; | |
2039 done = 1; | |
2040 | |
2041 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, | |
2042 &chroma_dc_coeff_token_len [0], 1, 1, | |
2043 &chroma_dc_coeff_token_bits[0], 1, 1); | |
2044 | |
2045 for(i=0; i<4; i++){ | |
2046 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, | |
2047 &coeff_token_len [i][0], 1, 1, | |
2048 &coeff_token_bits[i][0], 1, 1); | |
2049 } | |
2050 | |
2051 for(i=0; i<3; i++){ | |
2052 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4, | |
2053 &chroma_dc_total_zeros_len [i][0], 1, 1, | |
2054 &chroma_dc_total_zeros_bits[i][0], 1, 1); | |
2055 } | |
2056 for(i=0; i<15; i++){ | |
2057 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, | |
2058 &total_zeros_len [i][0], 1, 1, | |
2059 &total_zeros_bits[i][0], 1, 1); | |
2060 } | |
2061 | |
2062 for(i=0; i<6; i++){ | |
2063 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, | |
2064 &run_len [i][0], 1, 1, | |
2065 &run_bits[i][0], 1, 1); | |
2066 } | |
2067 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, | |
2068 &run_len [6][0], 1, 1, | |
2069 &run_bits[6][0], 1, 1); | |
2070 } | |
2071 } | |
2072 | |
2073 /** | |
2074 * Sets the intra prediction function pointers. | |
2075 */ | |
2076 static void init_pred_ptrs(H264Context *h){ | |
2077 // MpegEncContext * const s = &h->s; | |
2078 | |
2079 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c; | |
2080 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c; | |
2081 h->pred4x4[DC_PRED ]= pred4x4_dc_c; | |
2082 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c; | |
2083 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c; | |
2084 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c; | |
2085 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c; | |
2086 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c; | |
2087 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c; | |
2088 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c; | |
2089 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c; | |
2090 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c; | |
2091 | |
2092 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c; | |
2093 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c; | |
2094 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c; | |
2095 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c; | |
2096 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c; | |
2097 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c; | |
2098 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c; | |
2099 | |
2100 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c; | |
2101 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c; | |
2102 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c; | |
2103 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c; | |
2104 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c; | |
2105 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c; | |
2106 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c; | |
2107 } | |
2108 | |
2109 //FIXME factorize | |
2110 #define CHECKED_ALLOCZ(p, size)\ | |
2111 {\ | |
2112 p= av_mallocz(size);\ | |
2113 if(p==NULL){\ | |
2114 perror("malloc");\ | |
2115 goto fail;\ | |
2116 }\ | |
2117 } | |
2118 | |
2119 static void free_tables(H264Context *h){ | |
2120 av_freep(&h->intra4x4_pred_mode); | |
2121 av_freep(&h->non_zero_count); | |
2122 av_freep(&h->slice_table_base); | |
2123 h->slice_table= NULL; | |
2124 | |
2125 av_freep(&h->mb2b_xy); | |
2126 av_freep(&h->mb2b8_xy); | |
2127 } | |
2128 | |
2129 /** | |
2130 * allocates tables. | |
2131 * needs widzh/height | |
2132 */ | |
2133 static int alloc_tables(H264Context *h){ | |
2134 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
2135 const int big_mb_num= s->mb_stride * (s->mb_height+1); |
1168 | 2136 int x,y; |
2137 | |
2138 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t)) | |
2139 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t)) | |
2140 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t)) | |
2141 | |
2142 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t)); | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
2143 h->slice_table= h->slice_table_base + s->mb_stride + 1; |
1168 | 2144 |
2145 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t)); | |
2146 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t)); | |
2147 for(y=0; y<s->mb_height; y++){ | |
2148 for(x=0; x<s->mb_width; x++){ | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
2149 const int mb_xy= x + y*s->mb_stride; |
1168 | 2150 const int b_xy = 4*x + 4*y*h->b_stride; |
2151 const int b8_xy= 2*x + 2*y*h->b8_stride; | |
2152 | |
2153 h->mb2b_xy [mb_xy]= b_xy; | |
2154 h->mb2b8_xy[mb_xy]= b8_xy; | |
2155 } | |
2156 } | |
2157 | |
2158 return 0; | |
2159 fail: | |
2160 free_tables(h); | |
2161 return -1; | |
2162 } | |
2163 | |
2164 static void common_init(H264Context *h){ | |
2165 MpegEncContext * const s = &h->s; | |
2166 | |
2167 s->width = s->avctx->width; | |
2168 s->height = s->avctx->height; | |
2169 s->codec_id= s->avctx->codec->id; | |
2170 | |
2171 init_pred_ptrs(h); | |
2172 | |
2173 s->decode=1; //FIXME | |
2174 } | |
2175 | |
2176 static int decode_init(AVCodecContext *avctx){ | |
2177 H264Context *h= avctx->priv_data; | |
2178 MpegEncContext * const s = &h->s; | |
2179 | |
2180 s->avctx = avctx; | |
2181 common_init(h); | |
2182 | |
2183 s->out_format = FMT_H264; | |
2184 s->workaround_bugs= avctx->workaround_bugs; | |
2185 | |
2186 // set defaults | |
2187 s->progressive_sequence=1; | |
2188 // s->decode_mb= ff_h263_decode_mb; | |
2189 s->low_delay= 1; | |
2190 avctx->pix_fmt= PIX_FMT_YUV420P; | |
2191 | |
2192 decode_init_vlc(h); | |
2193 | |
2194 return 0; | |
2195 } | |
2196 | |
2197 static void frame_start(H264Context *h){ | |
2198 MpegEncContext * const s = &h->s; | |
2199 int i; | |
2200 | |
2201 MPV_frame_start(s, s->avctx); | |
2202 ff_er_frame_start(s); | |
2203 h->mmco_index=0; | |
2204 | |
2205 assert(s->linesize && s->uvlinesize); | |
2206 | |
2207 for(i=0; i<16; i++){ | |
2208 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3); | |
2209 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3); | |
2210 } | |
2211 for(i=0; i<4; i++){ | |
2212 h->block_offset[16+i]= | |
2213 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3); | |
2214 } | |
2215 | |
2216 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1; | |
2217 } | |
2218 | |
2219 static void hl_decode_mb(H264Context *h){ | |
2220 MpegEncContext * const s = &h->s; | |
2221 const int mb_x= s->mb_x; | |
2222 const int mb_y= s->mb_y; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
2223 const int mb_xy= mb_x + mb_y*s->mb_stride; |
1168 | 2224 const int mb_type= s->current_picture.mb_type[mb_xy]; |
2225 uint8_t *dest_y, *dest_cb, *dest_cr; | |
2226 int linesize, uvlinesize /*dct_offset*/; | |
2227 int i; | |
2228 | |
2229 if(!s->decode) | |
2230 return; | |
2231 | |
2232 if(s->mb_skiped){ | |
2233 } | |
2234 | |
2235 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16; | |
2236 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; | |
2237 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; | |
2238 | |
2239 if (h->mb_field_decoding_flag) { | |
2240 linesize = s->linesize * 2; | |
2241 uvlinesize = s->uvlinesize * 2; | |
2242 if(mb_y&1){ //FIXME move out of this func? | |
2243 dest_y -= s->linesize*15; | |
2244 dest_cb-= s->linesize*7; | |
2245 dest_cr-= s->linesize*7; | |
2246 } | |
2247 } else { | |
2248 linesize = s->linesize; | |
2249 uvlinesize = s->uvlinesize; | |
2250 // dct_offset = s->linesize * 16; | |
2251 } | |
2252 | |
2253 if(IS_INTRA(mb_type)){ | |
2254 if(!(s->flags&CODEC_FLAG_GRAY)){ | |
2255 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize); | |
2256 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize); | |
2257 } | |
2258 | |
2259 if(IS_INTRA4x4(mb_type)){ | |
2260 if(!s->encoding){ | |
2261 for(i=0; i<16; i++){ | |
2262 uint8_t * const ptr= dest_y + h->block_offset[i]; | |
2263 uint8_t *topright= ptr + 4 - linesize; | |
2264 const int topright_avail= (h->topright_samples_available<<i)&0x8000; | |
2265 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ]; | |
2266 int tr; | |
2267 | |
2268 if(!topright_avail){ | |
2269 tr= ptr[3 - linesize]*0x01010101; | |
2270 topright= (uint8_t*) &tr; | |
2271 } | |
2272 | |
2273 h->pred4x4[ dir ](ptr, topright, linesize); | |
1234 | 2274 if(h->non_zero_count_cache[ scan8[i] ]){ |
2275 if(s->codec_id == CODEC_ID_H264) | |
2276 h264_add_idct_c(ptr, h->mb + i*16, linesize); | |
2277 else | |
2278 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0); | |
2279 } | |
1168 | 2280 } |
2281 } | |
2282 }else{ | |
2283 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize); | |
1234 | 2284 if(s->codec_id == CODEC_ID_H264) |
2285 h264_luma_dc_dequant_idct_c(h->mb, s->qscale); | |
2286 else | |
2287 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale); | |
1168 | 2288 } |
1234 | 2289 }else if(s->codec_id == CODEC_ID_H264){ |
1168 | 2290 hl_motion(h, dest_y, dest_cb, dest_cr, |
2291 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, | |
2292 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab); | |
2293 } | |
2294 | |
2295 | |
2296 if(!IS_INTRA4x4(mb_type)){ | |
1250 | 2297 if(s->codec_id == CODEC_ID_H264){ |
2298 for(i=0; i<16; i++){ | |
2299 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below | |
2300 uint8_t * const ptr= dest_y + h->block_offset[i]; | |
1234 | 2301 h264_add_idct_c(ptr, h->mb + i*16, linesize); |
1250 | 2302 } |
2303 } | |
2304 }else{ | |
2305 for(i=0; i<16; i++){ | |
2306 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below | |
2307 uint8_t * const ptr= dest_y + h->block_offset[i]; | |
1234 | 2308 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0); |
1250 | 2309 } |
1168 | 2310 } |
2311 } | |
2312 } | |
2313 | |
2314 if(!(s->flags&CODEC_FLAG_GRAY)){ | |
2315 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp); | |
2316 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp); | |
1250 | 2317 if(s->codec_id == CODEC_ID_H264){ |
2318 for(i=16; i<16+4; i++){ | |
2319 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ | |
2320 uint8_t * const ptr= dest_cb + h->block_offset[i]; | |
1234 | 2321 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize); |
1250 | 2322 } |
2323 } | |
2324 for(i=20; i<20+4; i++){ | |
2325 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ | |
2326 uint8_t * const ptr= dest_cr + h->block_offset[i]; | |
2327 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize); | |
2328 } | |
1168 | 2329 } |
1250 | 2330 }else{ |
2331 for(i=16; i<16+4; i++){ | |
2332 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ | |
2333 uint8_t * const ptr= dest_cb + h->block_offset[i]; | |
1234 | 2334 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2); |
1250 | 2335 } |
2336 } | |
2337 for(i=20; i<20+4; i++){ | |
2338 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ | |
2339 uint8_t * const ptr= dest_cr + h->block_offset[i]; | |
2340 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2); | |
2341 } | |
1168 | 2342 } |
2343 } | |
2344 } | |
2345 } | |
2346 | |
2347 static void decode_mb_cabac(H264Context *h){ | |
2348 // MpegEncContext * const s = &h->s; | |
2349 } | |
2350 | |
2351 /** | |
2352 * fills the default_ref_list. | |
2353 */ | |
2354 static int fill_default_ref_list(H264Context *h){ | |
2355 MpegEncContext * const s = &h->s; | |
2356 int i; | |
2357 Picture sorted_short_ref[16]; | |
2358 | |
2359 if(h->slice_type==B_TYPE){ | |
2360 int out_i; | |
2361 int limit= -1; | |
2362 | |
2363 for(out_i=0; out_i<h->short_ref_count; out_i++){ | |
2364 int best_i=-1; | |
2365 int best_poc=-1; | |
2366 | |
2367 for(i=0; i<h->short_ref_count; i++){ | |
2368 const int poc= h->short_ref[i]->poc; | |
2369 if(poc > limit && poc < best_poc){ | |
2370 best_poc= poc; | |
2371 best_i= i; | |
2372 } | |
2373 } | |
2374 | |
2375 assert(best_i != -1); | |
2376 | |
2377 limit= best_poc; | |
2378 sorted_short_ref[out_i]= *h->short_ref[best_i]; | |
2379 } | |
2380 } | |
2381 | |
2382 if(s->picture_structure == PICT_FRAME){ | |
2383 if(h->slice_type==B_TYPE){ | |
2384 const int current_poc= s->current_picture_ptr->poc; | |
2385 int list; | |
2386 | |
2387 for(list=0; list<2; list++){ | |
2388 int index=0; | |
2389 | |
2390 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){ | |
2391 const int i2= list ? h->short_ref_count - i - 1 : i; | |
2392 const int poc= sorted_short_ref[i2].poc; | |
2393 | |
2394 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit | |
2395 | |
2396 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){ | |
2397 h->default_ref_list[list][index ]= sorted_short_ref[i2]; | |
2398 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num; | |
2399 } | |
2400 } | |
2401 | |
2402 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){ | |
2403 if(h->long_ref[i]->reference != 3) continue; | |
2404 | |
2405 h->default_ref_list[ list ][index ]= *h->long_ref[i]; | |
2406 h->default_ref_list[ list ][index++].pic_id= i;; | |
2407 } | |
2408 | |
2409 if(h->long_ref_count > 1 && h->short_ref_count==0){ | |
2410 Picture temp= h->default_ref_list[1][0]; | |
2411 h->default_ref_list[1][0] = h->default_ref_list[1][1]; | |
2412 h->default_ref_list[1][0] = temp; | |
2413 } | |
2414 | |
2415 if(index < h->ref_count[ list ]) | |
2416 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index)); | |
2417 } | |
2418 }else{ | |
2419 int index=0; | |
2420 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){ | |
2421 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit | |
2422 h->default_ref_list[0][index ]= *h->short_ref[i]; | |
2423 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num; | |
2424 } | |
2425 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){ | |
2426 if(h->long_ref[i]->reference != 3) continue; | |
2427 h->default_ref_list[0][index ]= *h->long_ref[i]; | |
2428 h->default_ref_list[0][index++].pic_id= i;; | |
2429 } | |
2430 if(index < h->ref_count[0]) | |
2431 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index)); | |
2432 } | |
2433 }else{ //FIELD | |
2434 if(h->slice_type==B_TYPE){ | |
2435 }else{ | |
2436 //FIXME second field balh | |
2437 } | |
2438 } | |
2439 return 0; | |
2440 } | |
2441 | |
2442 static int decode_ref_pic_list_reordering(H264Context *h){ | |
2443 MpegEncContext * const s = &h->s; | |
2444 int list; | |
2445 | |
2446 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func | |
2447 | |
2448 for(list=0; list<2; list++){ | |
2449 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]); | |
2450 | |
2451 if(get_bits1(&s->gb)){ | |
2452 int pred= h->curr_pic_num; | |
2453 int index; | |
2454 | |
2455 for(index=0; ; index++){ | |
2456 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb); | |
2457 int pic_id; | |
2458 int i; | |
2459 | |
2460 | |
2461 if(index >= h->ref_count[list]){ | |
2462 fprintf(stderr, "reference count overflow\n"); | |
2463 return -1; | |
2464 } | |
2465 | |
2466 if(reordering_of_pic_nums_idc<3){ | |
2467 if(reordering_of_pic_nums_idc<2){ | |
2468 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1; | |
2469 | |
2470 if(abs_diff_pic_num >= h->max_pic_num){ | |
2471 fprintf(stderr, "abs_diff_pic_num overflow\n"); | |
2472 return -1; | |
2473 } | |
2474 | |
2475 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num; | |
2476 else pred+= abs_diff_pic_num; | |
2477 pred &= h->max_pic_num - 1; | |
2478 | |
2479 for(i= h->ref_count[list]-1; i>=index; i--){ | |
2480 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0) | |
2481 break; | |
2482 } | |
2483 }else{ | |
2484 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx | |
2485 | |
2486 for(i= h->ref_count[list]-1; i>=index; i--){ | |
2487 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1) | |
2488 break; | |
2489 } | |
2490 } | |
2491 | |
2492 if(i < index){ | |
2493 fprintf(stderr, "reference picture missing during reorder\n"); | |
2494 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME | |
2495 }else if(i > index){ | |
2496 Picture tmp= h->ref_list[list][i]; | |
2497 for(; i>index; i--){ | |
2498 h->ref_list[list][i]= h->ref_list[list][i-1]; | |
2499 } | |
2500 h->ref_list[list][index]= tmp; | |
2501 } | |
2502 }else if(reordering_of_pic_nums_idc==3) | |
2503 break; | |
2504 else{ | |
2505 fprintf(stderr, "illegal reordering_of_pic_nums_idc\n"); | |
2506 return -1; | |
2507 } | |
2508 } | |
2509 } | |
2510 | |
2511 if(h->slice_type!=B_TYPE) break; | |
2512 } | |
2513 return 0; | |
2514 } | |
2515 | |
2516 static int pred_weight_table(H264Context *h){ | |
2517 MpegEncContext * const s = &h->s; | |
2518 int list, i; | |
2519 | |
2520 h->luma_log2_weight_denom= get_ue_golomb(&s->gb); | |
2521 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb); | |
2522 | |
2523 for(list=0; list<2; list++){ | |
2524 for(i=0; i<h->ref_count[list]; i++){ | |
2525 int luma_weight_flag, chroma_weight_flag; | |
2526 | |
2527 luma_weight_flag= get_bits1(&s->gb); | |
2528 if(luma_weight_flag){ | |
2529 h->luma_weight[list][i]= get_se_golomb(&s->gb); | |
2530 h->luma_offset[list][i]= get_se_golomb(&s->gb); | |
2531 } | |
2532 | |
2533 chroma_weight_flag= get_bits1(&s->gb); | |
2534 if(chroma_weight_flag){ | |
2535 int j; | |
2536 for(j=0; j<2; j++){ | |
2537 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb); | |
2538 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb); | |
2539 } | |
2540 } | |
2541 } | |
2542 if(h->slice_type != B_TYPE) break; | |
2543 } | |
2544 return 0; | |
2545 } | |
2546 | |
2547 /** | |
2548 * instantaneos decoder refresh. | |
2549 */ | |
2550 static void idr(H264Context *h){ | |
2551 int i; | |
2552 | |
2553 for(i=0; i<h->long_ref_count; i++){ | |
2554 h->long_ref[i]->reference=0; | |
2555 h->long_ref[i]= NULL; | |
2556 } | |
2557 h->long_ref_count=0; | |
2558 | |
2559 for(i=0; i<h->short_ref_count; i++){ | |
2560 h->short_ref[i]->reference=0; | |
2561 h->short_ref[i]= NULL; | |
2562 } | |
2563 h->short_ref_count=0; | |
2564 } | |
2565 | |
2566 /** | |
2567 * | |
2568 * @return the removed picture or NULL if an error occures | |
2569 */ | |
2570 static Picture * remove_short(H264Context *h, int frame_num){ | |
1169 | 2571 MpegEncContext * const s = &h->s; |
1168 | 2572 int i; |
2573 | |
1169 | 2574 if(s->avctx->debug&FF_DEBUG_MMCO) |
2575 printf("remove short %d count %d\n", frame_num, h->short_ref_count); | |
2576 | |
1168 | 2577 for(i=0; i<h->short_ref_count; i++){ |
2578 Picture *pic= h->short_ref[i]; | |
1169 | 2579 if(s->avctx->debug&FF_DEBUG_MMCO) |
1266 | 2580 printf("%d %d %p\n", i, pic->frame_num, pic); |
1168 | 2581 if(pic->frame_num == frame_num){ |
2582 h->short_ref[i]= NULL; | |
2583 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*)); | |
2584 h->short_ref_count--; | |
2585 return pic; | |
2586 } | |
2587 } | |
2588 return NULL; | |
2589 } | |
2590 | |
2591 /** | |
2592 * | |
2593 * @return the removed picture or NULL if an error occures | |
2594 */ | |
2595 static Picture * remove_long(H264Context *h, int i){ | |
2596 Picture *pic; | |
2597 | |
2598 if(i >= h->long_ref_count) return NULL; | |
2599 pic= h->long_ref[i]; | |
2600 if(pic==NULL) return NULL; | |
2601 | |
2602 h->long_ref[i]= NULL; | |
2603 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*)); | |
2604 h->long_ref_count--; | |
2605 | |
2606 return pic; | |
2607 } | |
2608 | |
2609 /** | |
2610 * Executes the reference picture marking (memory management control operations). | |
2611 */ | |
2612 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){ | |
2613 MpegEncContext * const s = &h->s; | |
2614 int i; | |
2615 int current_is_long=0; | |
2616 Picture *pic; | |
2617 | |
2618 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0) | |
2619 printf("no mmco here\n"); | |
2620 | |
2621 for(i=0; i<mmco_count; i++){ | |
2622 if(s->avctx->debug&FF_DEBUG_MMCO) | |
2623 printf("mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index); | |
2624 | |
2625 switch(mmco[i].opcode){ | |
2626 case MMCO_SHORT2UNUSED: | |
2627 pic= remove_short(h, mmco[i].short_frame_num); | |
2628 if(pic==NULL) return -1; | |
2629 pic->reference= 0; | |
2630 break; | |
2631 case MMCO_SHORT2LONG: | |
2632 pic= remove_long(h, mmco[i].long_index); | |
2633 if(pic) pic->reference=0; | |
2634 | |
2635 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num); | |
2636 h->long_ref[ mmco[i].long_index ]->long_ref=1; | |
2637 break; | |
2638 case MMCO_LONG2UNUSED: | |
2639 pic= remove_long(h, mmco[i].long_index); | |
2640 if(pic==NULL) return -1; | |
2641 pic->reference= 0; | |
2642 break; | |
2643 case MMCO_LONG: | |
2644 pic= remove_long(h, mmco[i].long_index); | |
2645 if(pic) pic->reference=0; | |
2646 | |
2647 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr; | |
2648 h->long_ref[ mmco[i].long_index ]->long_ref=1; | |
2649 h->long_ref_count++; | |
2650 | |
2651 current_is_long=1; | |
2652 break; | |
2653 case MMCO_SET_MAX_LONG: | |
2654 assert(mmco[i].long_index <= 16); | |
2655 while(mmco[i].long_index < h->long_ref_count){ | |
2656 pic= remove_long(h, mmco[i].long_index); | |
2657 pic->reference=0; | |
2658 } | |
2659 while(mmco[i].long_index > h->long_ref_count){ | |
2660 h->long_ref[ h->long_ref_count++ ]= NULL; | |
2661 } | |
2662 break; | |
2663 case MMCO_RESET: | |
2664 while(h->short_ref_count){ | |
2665 pic= remove_short(h, h->short_ref[0]->frame_num); | |
2666 pic->reference=0; | |
2667 } | |
2668 while(h->long_ref_count){ | |
2669 pic= remove_long(h, h->long_ref_count-1); | |
2670 pic->reference=0; | |
2671 } | |
2672 break; | |
2673 default: assert(0); | |
2674 } | |
2675 } | |
2676 | |
2677 if(!current_is_long){ | |
2678 pic= remove_short(h, s->current_picture_ptr->frame_num); | |
2679 if(pic){ | |
2680 pic->reference=0; | |
2681 fprintf(stderr, "illegal short term buffer state detected\n"); | |
2682 } | |
2683 | |
2684 if(h->short_ref_count) | |
1169 | 2685 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*)); |
2686 | |
2687 h->short_ref[0]= s->current_picture_ptr; | |
1168 | 2688 h->short_ref[0]->long_ref=0; |
2689 h->short_ref_count++; | |
2690 } | |
2691 | |
2692 return 0; | |
2693 } | |
2694 | |
2695 static int decode_ref_pic_marking(H264Context *h){ | |
2696 MpegEncContext * const s = &h->s; | |
2697 int i; | |
2698 | |
2699 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields | |
2700 s->broken_link= get_bits1(&s->gb) -1; | |
2701 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx | |
2702 if(h->mmco[0].long_index == -1) | |
2703 h->mmco_index= 0; | |
2704 else{ | |
2705 h->mmco[0].opcode= MMCO_LONG; | |
2706 h->mmco_index= 1; | |
2707 } | |
2708 }else{ | |
2709 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag | |
2710 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) { | |
2711 MMCOOpcode opcode= get_ue_golomb(&s->gb);; | |
2712 | |
2713 h->mmco[i].opcode= opcode; | |
2714 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){ | |
2715 h->mmco[i].short_frame_num= (h->frame_num - get_ue_golomb(&s->gb) - 1) & ((1<<h->sps.log2_max_frame_num)-1); //FIXME fields | |
2716 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){ | |
2717 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco); | |
2718 return -1; | |
2719 }*/ | |
2720 } | |
2721 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){ | |
2722 h->mmco[i].long_index= get_ue_golomb(&s->gb); | |
2723 if(/*h->mmco[i].long_index >= h->long_ref_count || h->long_ref[ h->mmco[i].long_index ] == NULL*/ h->mmco[i].long_index >= 16){ | |
2724 fprintf(stderr, "illegal long ref in memory management control operation %d\n", opcode); | |
2725 return -1; | |
2726 } | |
2727 } | |
2728 | |
2729 if(opcode > MMCO_LONG){ | |
2730 fprintf(stderr, "illegal memory management control operation %d\n", opcode); | |
2731 return -1; | |
2732 } | |
2733 } | |
2734 h->mmco_index= i; | |
2735 }else{ | |
2736 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count); | |
2737 | |
2738 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields | |
2739 h->mmco[0].opcode= MMCO_SHORT2UNUSED; | |
2740 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num; | |
2741 h->mmco_index= 1; | |
2742 }else | |
2743 h->mmco_index= 0; | |
2744 } | |
2745 } | |
2746 | |
2747 return 0; | |
2748 } | |
2749 | |
2750 static int init_poc(H264Context *h){ | |
2751 MpegEncContext * const s = &h->s; | |
2752 const int max_frame_num= 1<<h->sps.log2_max_frame_num; | |
2753 int field_poc[2]; | |
2754 | |
2755 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
2756 h->frame_num_offset= 0; | |
2757 }else{ | |
2758 if(h->frame_num < h->prev_frame_num) | |
2759 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num; | |
2760 else | |
2761 h->frame_num_offset= h->prev_frame_num_offset; | |
2762 } | |
2763 | |
2764 if(h->sps.poc_type==0){ | |
2765 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb; | |
2766 | |
2767 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2) | |
2768 h->poc_msb = h->prev_poc_msb + max_poc_lsb; | |
2769 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2) | |
2770 h->poc_msb = h->prev_poc_msb - max_poc_lsb; | |
2771 else | |
2772 h->poc_msb = h->prev_poc_msb; | |
2773 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb); | |
2774 field_poc[0] = | |
2775 field_poc[1] = h->poc_msb + h->poc_lsb; | |
2776 if(s->picture_structure == PICT_FRAME) | |
2777 field_poc[1] += h->delta_poc_bottom; | |
2778 }else if(h->sps.poc_type==1){ | |
2779 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc; | |
2780 int i; | |
2781 | |
2782 if(h->sps.poc_cycle_length != 0) | |
2783 abs_frame_num = h->frame_num_offset + h->frame_num; | |
2784 else | |
2785 abs_frame_num = 0; | |
2786 | |
2787 if(h->nal_ref_idc==0 && abs_frame_num > 0) | |
2788 abs_frame_num--; | |
2789 | |
2790 expected_delta_per_poc_cycle = 0; | |
2791 for(i=0; i < h->sps.poc_cycle_length; i++) | |
2792 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse | |
2793 | |
2794 if(abs_frame_num > 0){ | |
2795 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length; | |
2796 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length; | |
2797 | |
2798 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle; | |
2799 for(i = 0; i <= frame_num_in_poc_cycle; i++) | |
2800 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ]; | |
2801 } else | |
2802 expectedpoc = 0; | |
2803 | |
2804 if(h->nal_ref_idc == 0) | |
2805 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic; | |
2806 | |
2807 field_poc[0] = expectedpoc + h->delta_poc[0]; | |
2808 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field; | |
2809 | |
2810 if(s->picture_structure == PICT_FRAME) | |
2811 field_poc[1] += h->delta_poc[1]; | |
2812 }else{ | |
2813 int poc; | |
2814 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
2815 poc= 0; | |
2816 }else{ | |
2817 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num); | |
2818 else poc= 2*(h->frame_num_offset + h->frame_num) - 1; | |
2819 } | |
2820 field_poc[0]= poc; | |
2821 field_poc[1]= poc; | |
2822 } | |
2823 | |
2824 if(s->picture_structure != PICT_BOTTOM_FIELD) | |
2825 s->current_picture_ptr->field_poc[0]= field_poc[0]; | |
2826 if(s->picture_structure != PICT_TOP_FIELD) | |
2827 s->current_picture_ptr->field_poc[1]= field_poc[1]; | |
2828 if(s->picture_structure == PICT_FRAME) // FIXME field pix? | |
2829 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]); | |
2830 | |
2831 return 0; | |
2832 } | |
2833 | |
2834 /** | |
2835 * decodes a slice header. | |
2836 * this will allso call MPV_common_init() and frame_start() as needed | |
2837 */ | |
2838 static int decode_slice_header(H264Context *h){ | |
2839 MpegEncContext * const s = &h->s; | |
2840 int first_mb_in_slice, pps_id; | |
2841 int num_ref_idx_active_override_flag; | |
2842 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE}; | |
2843 float new_aspect; | |
2844 | |
2845 s->current_picture.reference= h->nal_ref_idc != 0; | |
2846 | |
2847 first_mb_in_slice= get_ue_golomb(&s->gb); | |
2848 | |
2849 h->slice_type= get_ue_golomb(&s->gb); | |
2850 if(h->slice_type > 9){ | |
2851 fprintf(stderr, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y); | |
2852 } | |
2853 if(h->slice_type > 4){ | |
2854 h->slice_type -= 5; | |
2855 h->slice_type_fixed=1; | |
2856 }else | |
2857 h->slice_type_fixed=0; | |
2858 | |
2859 h->slice_type= slice_type_map[ h->slice_type ]; | |
2860 | |
2861 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though | |
2862 | |
2863 pps_id= get_ue_golomb(&s->gb); | |
2864 if(pps_id>255){ | |
2865 fprintf(stderr, "pps_id out of range\n"); | |
2866 return -1; | |
2867 } | |
2868 h->pps= h->pps_buffer[pps_id]; | |
1174 | 2869 if(h->pps.slice_group_count == 0){ |
2870 fprintf(stderr, "non existing PPS referenced\n"); | |
2871 return -1; | |
2872 } | |
2873 | |
1168 | 2874 h->sps= h->sps_buffer[ h->pps.sps_id ]; |
1174 | 2875 if(h->sps.log2_max_frame_num == 0){ |
2876 fprintf(stderr, "non existing SPS referenced\n"); | |
2877 return -1; | |
2878 } | |
1168 | 2879 |
2880 s->mb_width= h->sps.mb_width; | |
2881 s->mb_height= h->sps.mb_height; | |
2882 | |
2883 h->b_stride= s->mb_width*4; | |
2884 h->b8_stride= s->mb_width*2; | |
2885 | |
2886 s->mb_x = first_mb_in_slice % s->mb_width; | |
2887 s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW | |
2888 | |
2889 s->width = 16*s->mb_width - 2*(h->pps.crop_left + h->pps.crop_right ); | |
2890 if(h->sps.frame_mbs_only_flag) | |
2891 s->height= 16*s->mb_height - 2*(h->pps.crop_top + h->pps.crop_bottom); | |
2892 else | |
2893 s->height= 16*s->mb_height - 4*(h->pps.crop_top + h->pps.crop_bottom); //FIXME recheck | |
2894 | |
2895 if(h->pps.crop_left || h->pps.crop_top){ | |
2896 fprintf(stderr, "insane croping not completly supported, this could look slightly wrong ...\n"); | |
2897 } | |
2898 | |
2899 if(s->aspected_height) //FIXME emms at end of slice ? | |
2900 new_aspect= h->sps.sar_width*s->width / (float)(s->height*h->sps.sar_height); | |
2901 else | |
2902 new_aspect=0; | |
2903 | |
2904 if (s->context_initialized | |
2905 && ( s->width != s->avctx->width || s->height != s->avctx->height | |
2906 || ABS(new_aspect - s->avctx->aspect_ratio) > 0.001)) { | |
2907 free_tables(h); | |
2908 MPV_common_end(s); | |
2909 } | |
2910 if (!s->context_initialized) { | |
2911 if (MPV_common_init(s) < 0) | |
2912 return -1; | |
2913 | |
2914 alloc_tables(h); | |
2915 | |
2916 s->avctx->width = s->width; | |
2917 s->avctx->height = s->height; | |
2918 s->avctx->aspect_ratio= new_aspect; | |
2919 } | |
2920 | |
2921 if(first_mb_in_slice == 0){ | |
2922 frame_start(h); | |
2923 } | |
2924 | |
1169 | 2925 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup |
1168 | 2926 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num); |
2927 | |
2928 if(h->sps.frame_mbs_only_flag){ | |
2929 s->picture_structure= PICT_FRAME; | |
2930 }else{ | |
2931 if(get_bits1(&s->gb)) //field_pic_flag | |
2932 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag | |
2933 else | |
2934 s->picture_structure= PICT_FRAME; | |
2935 } | |
2936 | |
2937 if(s->picture_structure==PICT_FRAME){ | |
2938 h->curr_pic_num= h->frame_num; | |
2939 h->max_pic_num= 1<< h->sps.log2_max_frame_num; | |
2940 }else{ | |
2941 h->curr_pic_num= 2*h->frame_num; | |
2942 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1); | |
2943 } | |
2944 | |
2945 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
2946 int idr_pic_id= get_ue_golomb(&s->gb); | |
2947 } | |
2948 | |
2949 if(h->sps.poc_type==0){ | |
2950 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb); | |
2951 | |
2952 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){ | |
2953 h->delta_poc_bottom= get_se_golomb(&s->gb); | |
2954 } | |
2955 } | |
2956 | |
2957 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){ | |
2958 h->delta_poc[0]= get_se_golomb(&s->gb); | |
2959 | |
2960 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME) | |
2961 h->delta_poc[1]= get_se_golomb(&s->gb); | |
2962 } | |
2963 | |
2964 init_poc(h); | |
2965 | |
2966 if(h->pps.redundant_pic_cnt_present){ | |
2967 h->redundant_pic_count= get_ue_golomb(&s->gb); | |
2968 } | |
2969 | |
2970 //set defaults, might be overriden a few line later | |
2971 h->ref_count[0]= h->pps.ref_count[0]; | |
2972 h->ref_count[1]= h->pps.ref_count[1]; | |
2973 | |
2974 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){ | |
2975 if(h->slice_type == B_TYPE){ | |
2976 h->direct_spatial_mv_pred= get_bits1(&s->gb); | |
2977 } | |
2978 num_ref_idx_active_override_flag= get_bits1(&s->gb); | |
2979 | |
2980 if(num_ref_idx_active_override_flag){ | |
2981 h->ref_count[0]= get_ue_golomb(&s->gb) + 1; | |
2982 if(h->slice_type==B_TYPE) | |
2983 h->ref_count[1]= get_ue_golomb(&s->gb) + 1; | |
2984 | |
2985 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){ | |
2986 fprintf(stderr, "reference overflow\n"); | |
2987 return -1; | |
2988 } | |
2989 } | |
2990 } | |
2991 | |
2992 if(first_mb_in_slice == 0){ | |
2993 fill_default_ref_list(h); | |
2994 } | |
2995 | |
2996 decode_ref_pic_list_reordering(h); | |
2997 | |
2998 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) | |
2999 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) ) | |
3000 pred_weight_table(h); | |
3001 | |
3002 if(s->current_picture.reference) | |
3003 decode_ref_pic_marking(h); | |
3004 //FIXME CABAC stuff | |
3005 | |
3006 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta | |
3007 //FIXME qscale / qp ... stuff | |
3008 if(h->slice_type == SP_TYPE){ | |
3009 int sp_for_switch_flag= get_bits1(&s->gb); | |
3010 } | |
3011 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){ | |
3012 int slice_qs_delta= get_se_golomb(&s->gb); | |
3013 } | |
3014 | |
3015 if( h->pps.deblocking_filter_parameters_present ) { | |
3016 h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb); | |
3017 if( h->disable_deblocking_filter_idc != 1 ) { | |
3018 h->slice_alpha_c0_offset_div2= get_se_golomb(&s->gb); | |
3019 h->slice_beta_offset_div2= get_se_golomb(&s->gb); | |
3020 } | |
3021 }else | |
3022 h->disable_deblocking_filter_idc= 0; | |
3023 | |
3024 #if 0 //FMO | |
3025 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5) | |
3026 slice_group_change_cycle= get_bits(&s->gb, ?); | |
3027 #endif | |
3028 | |
3029 if(s->avctx->debug&FF_DEBUG_PICT_INFO){ | |
3030 printf("mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n", | |
3031 first_mb_in_slice, | |
1264 | 3032 av_get_pict_type_char(h->slice_type), |
1168 | 3033 pps_id, h->frame_num, |
3034 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1], | |
3035 h->ref_count[0], h->ref_count[1], | |
3036 s->qscale, | |
3037 h->disable_deblocking_filter_idc | |
3038 ); | |
3039 } | |
3040 | |
3041 return 0; | |
3042 } | |
3043 | |
3044 /** | |
3045 * | |
3046 */ | |
3047 static inline int get_level_prefix(GetBitContext *gb){ | |
3048 unsigned int buf; | |
3049 int log; | |
3050 | |
3051 OPEN_READER(re, gb); | |
3052 UPDATE_CACHE(re, gb); | |
3053 buf=GET_CACHE(re, gb); | |
3054 | |
3055 log= 32 - av_log2(buf); | |
3056 #ifdef TRACE | |
3057 print_bin(buf>>(32-log), log); | |
3058 printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__); | |
3059 #endif | |
3060 | |
3061 LAST_SKIP_BITS(re, gb, log); | |
3062 CLOSE_READER(re, gb); | |
3063 | |
3064 return log-1; | |
3065 } | |
3066 | |
3067 /** | |
3068 * decodes a residual block. | |
3069 * @param n block index | |
3070 * @param scantable scantable | |
3071 * @param max_coeff number of coefficients in the block | |
3072 * @return <0 if an error occured | |
3073 */ | |
3074 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){ | |
3075 MpegEncContext * const s = &h->s; | |
3076 const uint16_t *qmul= dequant_coeff[qp]; | |
3077 static const int coeff_token_table_index[17]= {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3}; | |
3078 int level[16], run[16]; | |
3079 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones; | |
3080 | |
3081 //FIXME put trailing_onex into the context | |
3082 | |
3083 if(n == CHROMA_DC_BLOCK_INDEX){ | |
3084 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1); | |
3085 total_coeff= coeff_token>>2; | |
3086 }else{ | |
3087 if(n == LUMA_DC_BLOCK_INDEX){ | |
3088 total_coeff= pred_non_zero_count(h, 0); | |
3089 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2); | |
3090 total_coeff= coeff_token>>2; | |
3091 }else{ | |
3092 total_coeff= pred_non_zero_count(h, n); | |
3093 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2); | |
3094 total_coeff= coeff_token>>2; | |
3095 h->non_zero_count_cache[ scan8[n] ]= total_coeff; | |
3096 } | |
3097 } | |
3098 | |
3099 //FIXME set last_non_zero? | |
3100 | |
3101 if(total_coeff==0) | |
3102 return 0; | |
3103 | |
3104 trailing_ones= coeff_token&3; | |
1170 | 3105 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff); |
1168 | 3106 assert(total_coeff<=16); |
3107 | |
3108 for(i=0; i<trailing_ones; i++){ | |
3109 level[i]= 1 - 2*get_bits1(gb); | |
3110 } | |
3111 | |
3112 suffix_length= total_coeff > 10 && trailing_ones < 3; | |
3113 | |
3114 for(; i<total_coeff; i++){ | |
3115 const int prefix= get_level_prefix(gb); | |
3116 int level_code, mask; | |
3117 | |
3118 if(prefix<14){ //FIXME try to build a large unified VLC table for all this | |
3119 if(suffix_length) | |
3120 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part | |
3121 else | |
3122 level_code= (prefix<<suffix_length); //part | |
3123 }else if(prefix==14){ | |
3124 if(suffix_length) | |
3125 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part | |
3126 else | |
3127 level_code= prefix + get_bits(gb, 4); //part | |
3128 }else if(prefix==15){ | |
3129 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part | |
3130 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense | |
3131 }else{ | |
3132 fprintf(stderr, "prefix too large at %d %d\n", s->mb_x, s->mb_y); | |
3133 return -1; | |
3134 } | |
3135 | |
3136 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration | |
3137 | |
3138 mask= -(level_code&1); | |
3139 level[i]= (((2+level_code)>>1) ^ mask) - mask; | |
3140 | |
3141 if(suffix_length==0) suffix_length=1; //FIXME split first iteration | |
3142 | |
3143 #if 1 | |
3144 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++; | |
3145 #else | |
3146 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++; | |
3147 ? == prefix > 2 or sth | |
3148 #endif | |
1170 | 3149 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length); |
1168 | 3150 } |
3151 | |
3152 if(total_coeff == max_coeff) | |
3153 zeros_left=0; | |
3154 else{ | |
3155 if(n == CHROMA_DC_BLOCK_INDEX) | |
3156 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1); | |
3157 else | |
3158 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1); | |
3159 } | |
3160 | |
3161 for(i=0; i<total_coeff-1; i++){ | |
3162 if(zeros_left <=0) | |
3163 break; | |
3164 else if(zeros_left < 7){ | |
3165 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1); | |
3166 }else{ | |
3167 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); | |
3168 } | |
3169 zeros_left -= run[i]; | |
3170 } | |
3171 | |
3172 if(zeros_left<0){ | |
3173 fprintf(stderr, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y); | |
3174 return -1; | |
3175 } | |
3176 | |
3177 for(; i<total_coeff-1; i++){ | |
3178 run[i]= 0; | |
3179 } | |
3180 | |
3181 run[i]= zeros_left; | |
3182 | |
3183 coeff_num=-1; | |
3184 if(n > 24){ | |
3185 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode? | |
3186 int j; | |
3187 | |
3188 coeff_num += run[i] + 1; //FIXME add 1 earlier ? | |
3189 j= scantable[ coeff_num ]; | |
3190 | |
3191 block[j]= level[i]; | |
3192 } | |
3193 }else{ | |
3194 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode? | |
3195 int j; | |
3196 | |
3197 coeff_num += run[i] + 1; //FIXME add 1 earlier ? | |
3198 j= scantable[ coeff_num ]; | |
3199 | |
3200 block[j]= level[i] * qmul[j]; | |
3201 // printf("%d %d ", block[j], qmul[j]); | |
3202 } | |
3203 } | |
3204 return 0; | |
3205 } | |
3206 | |
3207 /** | |
3208 * decodes a macroblock | |
3209 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed | |
3210 */ | |
3211 static int decode_mb(H264Context *h){ | |
3212 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
3213 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1169 | 3214 int mb_type, partition_count, cbp; |
1168 | 3215 |
1252 | 3216 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong? |
1168 | 3217 |
1170 | 3218 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y); |
1168 | 3219 |
3220 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){ | |
3221 if(s->mb_skip_run==-1) | |
3222 s->mb_skip_run= get_ue_golomb(&s->gb); | |
3223 | |
3224 if (s->mb_skip_run--) { | |
1187 | 3225 int mx, my; |
1168 | 3226 /* skip mb */ |
3227 //FIXME b frame | |
3228 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0; | |
3229 | |
3230 memset(h->non_zero_count[mb_xy], 0, 16); | |
3231 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui | |
3232 | |
3233 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){ | |
3234 h->mb_field_decoding_flag= get_bits1(&s->gb); | |
3235 } | |
3236 | |
3237 if(h->mb_field_decoding_flag) | |
3238 mb_type|= MB_TYPE_INTERLACED; | |
3239 | |
3240 fill_caches(h, mb_type); //FIXME check what is needed and what not ... | |
3241 pred_pskip_motion(h, &mx, &my); | |
3242 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1); | |
1269 | 3243 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4); |
1168 | 3244 write_back_motion(h, mb_type); |
3245 | |
3246 s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type | |
3247 h->slice_table[ mb_xy ]= h->slice_num; | |
3248 | |
3249 h->prev_mb_skiped= 1; | |
3250 return 0; | |
3251 } | |
3252 } | |
3253 if(h->sps.mb_aff /* && !field pic FIXME needed? */){ | |
3254 if((s->mb_y&1)==0) | |
3255 h->mb_field_decoding_flag = get_bits1(&s->gb); | |
3256 }else | |
3257 h->mb_field_decoding_flag=0; //FIXME som ed note ?! | |
3258 | |
3259 h->prev_mb_skiped= 0; | |
3260 | |
3261 mb_type= get_ue_golomb(&s->gb); | |
3262 if(h->slice_type == B_TYPE){ | |
3263 if(mb_type < 23){ | |
3264 partition_count= b_mb_type_info[mb_type].partition_count; | |
3265 mb_type= b_mb_type_info[mb_type].type; | |
3266 }else{ | |
3267 mb_type -= 23; | |
3268 goto decode_intra_mb; | |
3269 } | |
3270 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){ | |
3271 if(mb_type < 5){ | |
3272 partition_count= p_mb_type_info[mb_type].partition_count; | |
3273 mb_type= p_mb_type_info[mb_type].type; | |
3274 }else{ | |
3275 mb_type -= 5; | |
3276 goto decode_intra_mb; | |
3277 } | |
3278 }else{ | |
3279 assert(h->slice_type == I_TYPE); | |
3280 decode_intra_mb: | |
3281 if(mb_type > 25){ | |
1264 | 3282 fprintf(stderr, "mb_type %d in %c slice to large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y); |
1168 | 3283 return -1; |
3284 } | |
3285 partition_count=0; | |
3286 cbp= i_mb_type_info[mb_type].cbp; | |
3287 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode; | |
3288 mb_type= i_mb_type_info[mb_type].type; | |
3289 } | |
3290 | |
3291 if(h->mb_field_decoding_flag) | |
3292 mb_type |= MB_TYPE_INTERLACED; | |
3293 | |
3294 s->current_picture.mb_type[mb_xy]= mb_type; | |
3295 h->slice_table[ mb_xy ]= h->slice_num; | |
3296 | |
3297 if(IS_INTRA_PCM(mb_type)){ | |
3298 const uint8_t *ptr; | |
1187 | 3299 int x, y; |
1168 | 3300 |
3301 // we assume these blocks are very rare so we dont optimize it | |
3302 align_get_bits(&s->gb); | |
3303 | |
3304 ptr= s->gb.buffer + get_bits_count(&s->gb); | |
3305 | |
3306 for(y=0; y<16; y++){ | |
3307 const int index= 4*(y&3) + 64*(y>>2); | |
3308 for(x=0; x<16; x++){ | |
3309 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++); | |
3310 } | |
3311 } | |
3312 for(y=0; y<8; y++){ | |
3313 const int index= 256 + 4*(y&3) + 32*(y>>2); | |
3314 for(x=0; x<8; x++){ | |
3315 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++); | |
3316 } | |
3317 } | |
3318 for(y=0; y<8; y++){ | |
3319 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2); | |
3320 for(x=0; x<8; x++){ | |
3321 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++); | |
3322 } | |
3323 } | |
3324 | |
3325 skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers | |
3326 | |
3327 memset(h->non_zero_count[mb_xy], 16, 16); | |
3328 | |
3329 return 0; | |
3330 } | |
3331 | |
3332 fill_caches(h, mb_type); | |
3333 | |
3334 //mb_pred | |
3335 if(IS_INTRA(mb_type)){ | |
3336 // init_top_left_availability(h); | |
3337 if(IS_INTRA4x4(mb_type)){ | |
3338 int i; | |
3339 | |
3340 // fill_intra4x4_pred_table(h); | |
3341 for(i=0; i<16; i++){ | |
3342 const int mode_coded= !get_bits1(&s->gb); | |
3343 const int predicted_mode= pred_intra_mode(h, i); | |
3344 int mode; | |
3345 | |
3346 if(mode_coded){ | |
3347 const int rem_mode= get_bits(&s->gb, 3); | |
3348 if(rem_mode<predicted_mode) | |
3349 mode= rem_mode; | |
3350 else | |
3351 mode= rem_mode + 1; | |
3352 }else{ | |
3353 mode= predicted_mode; | |
3354 } | |
3355 | |
3356 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode; | |
3357 } | |
3358 write_back_intra_pred_mode(h); | |
3359 if( check_intra4x4_pred_mode(h) < 0) | |
3360 return -1; | |
3361 }else{ | |
3362 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode); | |
3363 if(h->intra16x16_pred_mode < 0) | |
3364 return -1; | |
3365 } | |
3366 h->chroma_pred_mode= get_ue_golomb(&s->gb); | |
3367 | |
3368 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode); | |
3369 if(h->chroma_pred_mode < 0) | |
3370 return -1; | |
3371 }else if(partition_count==4){ | |
3372 int i, j, sub_partition_count[4], list, ref[2][4]; | |
3373 | |
3374 if(h->slice_type == B_TYPE){ | |
3375 for(i=0; i<4; i++){ | |
3376 h->sub_mb_type[i]= get_ue_golomb(&s->gb); | |
3377 if(h->sub_mb_type[i] >=13){ | |
3378 fprintf(stderr, "B sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y); | |
3379 return -1; | |
3380 } | |
3381 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; | |
3382 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type; | |
3383 } | |
3384 }else{ | |
3385 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ? | |
3386 for(i=0; i<4; i++){ | |
3387 h->sub_mb_type[i]= get_ue_golomb(&s->gb); | |
3388 if(h->sub_mb_type[i] >=4){ | |
3389 fprintf(stderr, "P sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y); | |
3390 return -1; | |
3391 } | |
3392 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; | |
3393 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type; | |
3394 } | |
3395 } | |
3396 | |
3397 for(list=0; list<2; list++){ | |
3398 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list]; | |
3399 if(ref_count == 0) continue; | |
3400 for(i=0; i<4; i++){ | |
3401 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){ | |
3402 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip? | |
3403 }else{ | |
3404 //FIXME | |
3405 ref[list][i] = -1; | |
3406 } | |
3407 } | |
3408 } | |
3409 | |
3410 for(list=0; list<2; list++){ | |
3411 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list]; | |
3412 if(ref_count == 0) continue; | |
3413 | |
3414 for(i=0; i<4; i++){ | |
3415 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]= | |
3416 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i]; | |
3417 | |
3418 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){ | |
3419 const int sub_mb_type= h->sub_mb_type[i]; | |
3420 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1; | |
3421 for(j=0; j<sub_partition_count[i]; j++){ | |
3422 int mx, my; | |
3423 const int index= 4*i + block_width*j; | |
3424 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ]; | |
3425 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my); | |
3426 mx += get_se_golomb(&s->gb); | |
3427 my += get_se_golomb(&s->gb); | |
1170 | 3428 tprintf("final mv:%d %d\n", mx, my); |
3429 | |
1168 | 3430 if(IS_SUB_8X8(sub_mb_type)){ |
3431 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= | |
3432 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx; | |
3433 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= | |
3434 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my; | |
3435 }else if(IS_SUB_8X4(sub_mb_type)){ | |
3436 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx; | |
3437 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my; | |
3438 }else if(IS_SUB_4X8(sub_mb_type)){ | |
3439 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx; | |
3440 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my; | |
3441 }else{ | |
3442 assert(IS_SUB_4X4(sub_mb_type)); | |
3443 mv_cache[ 0 ][0]= mx; | |
3444 mv_cache[ 0 ][1]= my; | |
3445 } | |
3446 } | |
3447 }else{ | |
3448 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0]; | |
3449 p[0] = p[1]= | |
3450 p[8] = p[9]= 0; | |
3451 } | |
3452 } | |
3453 } | |
3454 }else if(!IS_DIRECT(mb_type)){ | |
3455 int list, mx, my, i; | |
3456 //FIXME we should set ref_idx_l? to 0 if we use that later ... | |
3457 if(IS_16X16(mb_type)){ | |
3458 for(list=0; list<2; list++){ | |
3459 if(h->ref_count[0]>0){ | |
3460 if(IS_DIR(mb_type, 0, list)){ | |
3461 const int val= get_te0_golomb(&s->gb, h->ref_count[list]); | |
3462 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1); | |
3463 } | |
3464 } | |
3465 } | |
3466 for(list=0; list<2; list++){ | |
3467 if(IS_DIR(mb_type, 0, list)){ | |
3468 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my); | |
3469 mx += get_se_golomb(&s->gb); | |
3470 my += get_se_golomb(&s->gb); | |
1170 | 3471 tprintf("final mv:%d %d\n", mx, my); |
3472 | |
1269 | 3473 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4); |
1168 | 3474 } |
3475 } | |
3476 } | |
3477 else if(IS_16X8(mb_type)){ | |
3478 for(list=0; list<2; list++){ | |
3479 if(h->ref_count[list]>0){ | |
3480 for(i=0; i<2; i++){ | |
3481 if(IS_DIR(mb_type, i, list)){ | |
3482 const int val= get_te0_golomb(&s->gb, h->ref_count[list]); | |
3483 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1); | |
3484 } | |
3485 } | |
3486 } | |
3487 } | |
3488 for(list=0; list<2; list++){ | |
3489 for(i=0; i<2; i++){ | |
3490 if(IS_DIR(mb_type, i, list)){ | |
3491 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my); | |
3492 mx += get_se_golomb(&s->gb); | |
3493 my += get_se_golomb(&s->gb); | |
1170 | 3494 tprintf("final mv:%d %d\n", mx, my); |
3495 | |
1269 | 3496 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4); |
1168 | 3497 } |
3498 } | |
3499 } | |
3500 }else{ | |
3501 assert(IS_8X16(mb_type)); | |
3502 for(list=0; list<2; list++){ | |
3503 if(h->ref_count[list]>0){ | |
3504 for(i=0; i<2; i++){ | |
3505 if(IS_DIR(mb_type, i, list)){ //FIXME optimize | |
3506 const int val= get_te0_golomb(&s->gb, h->ref_count[list]); | |
3507 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1); | |
3508 } | |
3509 } | |
3510 } | |
3511 } | |
3512 for(list=0; list<2; list++){ | |
3513 for(i=0; i<2; i++){ | |
3514 if(IS_DIR(mb_type, i, list)){ | |
3515 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my); | |
3516 mx += get_se_golomb(&s->gb); | |
3517 my += get_se_golomb(&s->gb); | |
1170 | 3518 tprintf("final mv:%d %d\n", mx, my); |
3519 | |
1269 | 3520 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4); |
1168 | 3521 } |
3522 } | |
3523 } | |
3524 } | |
3525 } | |
3526 | |
3527 if(IS_INTER(mb_type)) | |
3528 write_back_motion(h, mb_type); | |
3529 | |
3530 if(!IS_INTRA16x16(mb_type)){ | |
3531 cbp= get_ue_golomb(&s->gb); | |
3532 if(cbp > 47){ | |
3533 fprintf(stderr, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y); | |
3534 return -1; | |
3535 } | |
3536 | |
3537 if(IS_INTRA4x4(mb_type)) | |
3538 cbp= golomb_to_intra4x4_cbp[cbp]; | |
3539 else | |
3540 cbp= golomb_to_inter_cbp[cbp]; | |
3541 } | |
3542 | |
3543 if(cbp || IS_INTRA16x16(mb_type)){ | |
3544 int i8x8, i4x4, chroma_idx; | |
3545 int chroma_qp, dquant; | |
3546 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr; | |
3547 const uint8_t *scan, *dc_scan; | |
3548 | |
3549 // fill_non_zero_count_cache(h); | |
3550 | |
3551 if(IS_INTERLACED(mb_type)){ | |
3552 scan= field_scan; | |
3553 dc_scan= luma_dc_field_scan; | |
3554 }else{ | |
3555 scan= zigzag_scan; | |
3556 dc_scan= luma_dc_zigzag_scan; | |
3557 } | |
3558 | |
3559 dquant= get_se_golomb(&s->gb); | |
3560 | |
3561 if( dquant > 25 || dquant < -26 ){ | |
3562 fprintf(stderr, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y); | |
3563 return -1; | |
3564 } | |
3565 | |
3566 s->qscale += dquant; | |
3567 if(((unsigned)s->qscale) > 51){ | |
3568 if(s->qscale<0) s->qscale+= 52; | |
3569 else s->qscale-= 52; | |
3570 } | |
3571 | |
3572 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale); | |
3573 if(IS_INTRA16x16(mb_type)){ | |
3574 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){ | |
3575 return -1; //FIXME continue if partotioned and other retirn -1 too | |
3576 } | |
3577 | |
3578 assert((cbp&15) == 0 || (cbp&15) == 15); | |
3579 | |
3580 if(cbp&15){ | |
3581 for(i8x8=0; i8x8<4; i8x8++){ | |
3582 for(i4x4=0; i4x4<4; i4x4++){ | |
3583 const int index= i4x4 + 4*i8x8; | |
3584 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){ | |
3585 return -1; | |
3586 } | |
3587 } | |
3588 } | |
3589 }else{ | |
3590 memset(&h->non_zero_count_cache[8], 0, 8*4); //FIXME stupid & slow | |
3591 } | |
3592 }else{ | |
3593 for(i8x8=0; i8x8<4; i8x8++){ | |
3594 if(cbp & (1<<i8x8)){ | |
3595 for(i4x4=0; i4x4<4; i4x4++){ | |
3596 const int index= i4x4 + 4*i8x8; | |
3597 | |
3598 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){ | |
3599 return -1; | |
3600 } | |
3601 } | |
3602 }else{ | |
3603 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ]; | |
3604 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0; | |
3605 } | |
3606 } | |
3607 } | |
3608 | |
3609 if(cbp&0x30){ | |
3610 for(chroma_idx=0; chroma_idx<2; chroma_idx++) | |
3611 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){ | |
3612 return -1; | |
3613 } | |
3614 } | |
3615 | |
3616 if(cbp&0x20){ | |
3617 for(chroma_idx=0; chroma_idx<2; chroma_idx++){ | |
3618 for(i4x4=0; i4x4<4; i4x4++){ | |
3619 const int index= 16 + 4*chroma_idx + i4x4; | |
3620 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){ | |
3621 return -1; | |
3622 } | |
3623 } | |
3624 } | |
3625 }else{ | |
3626 uint8_t * const nnz= &h->non_zero_count_cache[0]; | |
3627 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] = | |
3628 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0; | |
3629 } | |
3630 }else{ | |
3631 memset(&h->non_zero_count_cache[8], 0, 8*5); | |
3632 } | |
3633 write_back_non_zero_count(h); | |
3634 | |
3635 return 0; | |
3636 } | |
3637 | |
3638 static int decode_slice(H264Context *h){ | |
3639 MpegEncContext * const s = &h->s; | |
3640 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F; | |
3641 | |
3642 s->mb_skip_run= -1; | |
3643 | |
3644 #if 1 | |
3645 for(;;){ | |
3646 int ret= decode_mb(h); | |
3647 | |
3648 hl_decode_mb(h); | |
3649 | |
3650 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ? | |
3651 s->mb_y++; | |
3652 ret= decode_mb(h); | |
3653 | |
3654 hl_decode_mb(h); | |
3655 s->mb_y--; | |
3656 } | |
3657 | |
3658 if(ret<0){ | |
3659 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y); | |
3660 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); | |
3661 | |
3662 return -1; | |
3663 } | |
3664 | |
3665 if(++s->mb_x >= s->mb_width){ | |
3666 s->mb_x=0; | |
3667 ff_draw_horiz_band(s, 16*s->mb_y, 16); | |
3668 if(++s->mb_y >= s->mb_height){ | |
1170 | 3669 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits); |
3670 | |
1168 | 3671 if(get_bits_count(&s->gb) == s->gb.size_in_bits){ |
3672 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); | |
3673 | |
3674 return 0; | |
3675 }else{ | |
3676 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); | |
3677 | |
3678 return -1; | |
3679 } | |
3680 } | |
3681 } | |
3682 | |
3683 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){ | |
3684 if(get_bits_count(&s->gb) == s->gb.size_in_bits){ | |
3685 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); | |
3686 | |
3687 return 0; | |
3688 }else{ | |
3689 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); | |
3690 | |
3691 return -1; | |
3692 } | |
3693 } | |
3694 } | |
3695 #endif | |
3696 #if 0 | |
3697 for(;s->mb_y < s->mb_height; s->mb_y++){ | |
3698 for(;s->mb_x < s->mb_width; s->mb_x++){ | |
3699 int ret= decode_mb(h); | |
3700 | |
3701 hl_decode_mb(h); | |
3702 | |
3703 if(ret<0){ | |
3704 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y); | |
3705 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); | |
3706 | |
3707 return -1; | |
3708 } | |
3709 | |
3710 if(++s->mb_x >= s->mb_width){ | |
3711 s->mb_x=0; | |
3712 if(++s->mb_y >= s->mb_height){ | |
3713 if(get_bits_count(s->gb) == s->gb.size_in_bits){ | |
3714 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); | |
3715 | |
3716 return 0; | |
3717 }else{ | |
3718 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); | |
3719 | |
3720 return -1; | |
3721 } | |
3722 } | |
3723 } | |
3724 | |
3725 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){ | |
3726 if(get_bits_count(s->gb) == s->gb.size_in_bits){ | |
3727 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); | |
3728 | |
3729 return 0; | |
3730 }else{ | |
3731 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); | |
3732 | |
3733 return -1; | |
3734 } | |
3735 } | |
3736 } | |
3737 s->mb_x=0; | |
3738 ff_draw_horiz_band(s, 16*s->mb_y, 16); | |
3739 } | |
3740 #endif | |
3741 return -1; //not reached | |
3742 } | |
3743 | |
3744 static inline int decode_vui_parameters(H264Context *h, SPS *sps){ | |
3745 MpegEncContext * const s = &h->s; | |
3746 int aspect_ratio_info_present_flag, aspect_ratio_idc; | |
3747 | |
3748 aspect_ratio_info_present_flag= get_bits1(&s->gb); | |
3749 | |
3750 if( aspect_ratio_info_present_flag ) { | |
3751 aspect_ratio_idc= get_bits(&s->gb, 8); | |
3752 if( aspect_ratio_idc == EXTENDED_SAR ) { | |
3753 sps->sar_width= get_bits(&s->gb, 16); | |
3754 sps->sar_height= get_bits(&s->gb, 16); | |
3755 }else if(aspect_ratio_idc < 16){ | |
3756 sps->sar_width= pixel_aspect[aspect_ratio_idc][0]; | |
3757 sps->sar_height= pixel_aspect[aspect_ratio_idc][1]; | |
3758 }else{ | |
3759 fprintf(stderr, "illegal aspect ratio\n"); | |
3760 return -1; | |
3761 } | |
3762 }else{ | |
3763 sps->sar_width= | |
3764 sps->sar_height= 0; | |
3765 } | |
3766 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height); | |
3767 #if 0 | |
3768 | overscan_info_present_flag |0 |u(1) | | |
3769 | if( overscan_info_present_flag ) | | | | |
3770 | overscan_appropriate_flag |0 |u(1) | | |
3771 | video_signal_type_present_flag |0 |u(1) | | |
3772 | if( video_signal_type_present_flag ) { | | | | |
3773 | video_format |0 |u(3) | | |
3774 | video_full_range_flag |0 |u(1) | | |
3775 | colour_description_present_flag |0 |u(1) | | |
3776 | if( colour_description_present_flag ) { | | | | |
3777 | colour_primaries |0 |u(8) | | |
3778 | transfer_characteristics |0 |u(8) | | |
3779 | matrix_coefficients |0 |u(8) | | |
3780 | } | | | | |
3781 | } | | | | |
3782 | chroma_location_info_present_flag |0 |u(1) | | |
3783 | if ( chroma_location_info_present_flag ) { | | | | |
3784 | chroma_sample_location_type_top_field |0 |ue(v) | | |
3785 | chroma_sample_location_type_bottom_field |0 |ue(v) | | |
3786 | } | | | | |
3787 | timing_info_present_flag |0 |u(1) | | |
3788 | if( timing_info_present_flag ) { | | | | |
3789 | num_units_in_tick |0 |u(32) | | |
3790 | time_scale |0 |u(32) | | |
3791 | fixed_frame_rate_flag |0 |u(1) | | |
3792 | } | | | | |
3793 | nal_hrd_parameters_present_flag |0 |u(1) | | |
3794 | if( nal_hrd_parameters_present_flag = = 1) | | | | |
3795 | hrd_parameters( ) | | | | |
3796 | vcl_hrd_parameters_present_flag |0 |u(1) | | |
3797 | if( vcl_hrd_parameters_present_flag = = 1) | | | | |
3798 | hrd_parameters( ) | | | | |
3799 | if( ( nal_hrd_parameters_present_flag = = 1 | || | | | |
3800 | | | | | |
3801 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | | | |
3802 | low_delay_hrd_flag |0 |u(1) | | |
3803 | bitstream_restriction_flag |0 |u(1) | | |
3804 | if( bitstream_restriction_flag ) { |0 |u(1) | | |
3805 | motion_vectors_over_pic_boundaries_flag |0 |u(1) | | |
3806 | max_bytes_per_pic_denom |0 |ue(v) | | |
3807 | max_bits_per_mb_denom |0 |ue(v) | | |
3808 | log2_max_mv_length_horizontal |0 |ue(v) | | |
3809 | log2_max_mv_length_vertical |0 |ue(v) | | |
3810 | num_reorder_frames |0 |ue(v) | | |
3811 | max_dec_frame_buffering |0 |ue(v) | | |
3812 | } | | | | |
3813 |} | | | | |
3814 #endif | |
3815 return 0; | |
3816 } | |
3817 | |
3818 static inline int decode_seq_parameter_set(H264Context *h){ | |
3819 MpegEncContext * const s = &h->s; | |
3820 int profile_idc, level_idc, multiple_slice_groups, arbitrary_slice_order, redundant_slices; | |
3821 int sps_id, i; | |
3822 SPS *sps; | |
3823 | |
3824 profile_idc= get_bits(&s->gb, 8); | |
3825 level_idc= get_bits(&s->gb, 8); | |
3826 multiple_slice_groups= get_bits1(&s->gb); | |
3827 arbitrary_slice_order= get_bits1(&s->gb); | |
3828 redundant_slices= get_bits1(&s->gb); | |
3829 | |
3830 sps_id= get_ue_golomb(&s->gb); | |
3831 | |
3832 sps= &h->sps_buffer[ sps_id ]; | |
3833 | |
3834 sps->profile_idc= profile_idc; | |
3835 sps->level_idc= level_idc; | |
3836 sps->multiple_slice_groups= multiple_slice_groups; | |
3837 sps->arbitrary_slice_order= arbitrary_slice_order; | |
3838 sps->redundant_slices= redundant_slices; | |
3839 | |
3840 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4; | |
3841 | |
3842 sps->poc_type= get_ue_golomb(&s->gb); | |
3843 | |
3844 if(sps->poc_type == 0){ //FIXME #define | |
3845 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4; | |
3846 } else if(sps->poc_type == 1){//FIXME #define | |
3847 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb); | |
3848 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb); | |
3849 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb); | |
3850 sps->poc_cycle_length= get_ue_golomb(&s->gb); | |
3851 | |
3852 for(i=0; i<sps->poc_cycle_length; i++) | |
3853 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb); | |
3854 } | |
3855 if(sps->poc_type > 2){ | |
3856 fprintf(stderr, "illegal POC type %d\n", sps->poc_type); | |
3857 return -1; | |
3858 } | |
3859 | |
3860 sps->ref_frame_count= get_ue_golomb(&s->gb); | |
3861 sps->required_frame_num_update_behaviour_flag= get_bits1(&s->gb); | |
3862 sps->mb_width= get_ue_golomb(&s->gb) + 1; | |
3863 sps->mb_height= get_ue_golomb(&s->gb) + 1; | |
3864 sps->frame_mbs_only_flag= get_bits1(&s->gb); | |
3865 if(!sps->frame_mbs_only_flag) | |
3866 sps->mb_aff= get_bits1(&s->gb); | |
3867 else | |
3868 sps->mb_aff= 0; | |
3869 | |
3870 sps->direct_8x8_inference_flag= get_bits1(&s->gb); | |
3871 | |
3872 sps->vui_parameters_present_flag= get_bits1(&s->gb); | |
3873 if( sps->vui_parameters_present_flag ) | |
3874 decode_vui_parameters(h, sps); | |
3875 | |
3876 if(s->avctx->debug&FF_DEBUG_PICT_INFO){ | |
3877 printf("sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s %s\n", | |
3878 sps_id, sps->profile_idc, sps->level_idc, | |
3879 sps->poc_type, | |
3880 sps->ref_frame_count, | |
3881 sps->mb_width, sps->mb_height, | |
3882 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"), | |
3883 sps->direct_8x8_inference_flag ? "8B8" : "", | |
3884 sps->vui_parameters_present_flag ? "VUI" : "" | |
3885 ); | |
3886 } | |
3887 return 0; | |
3888 } | |
3889 | |
3890 static inline int decode_picture_parameter_set(H264Context *h){ | |
3891 MpegEncContext * const s = &h->s; | |
3892 int pps_id= get_ue_golomb(&s->gb); | |
3893 PPS *pps= &h->pps_buffer[pps_id]; | |
3894 | |
3895 pps->sps_id= get_ue_golomb(&s->gb); | |
3896 pps->cabac= get_bits1(&s->gb); | |
3897 pps->pic_order_present= get_bits1(&s->gb); | |
3898 pps->slice_group_count= get_ue_golomb(&s->gb) + 1; | |
3899 if(pps->slice_group_count > 1 ){ | |
3900 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb); | |
3901 fprintf(stderr, "FMO not supported\n"); | |
3902 switch(pps->mb_slice_group_map_type){ | |
3903 case 0: | |
3904 #if 0 | |
3905 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | | | |
3906 | run_length[ i ] |1 |ue(v) | | |
3907 #endif | |
3908 break; | |
3909 case 2: | |
3910 #if 0 | |
3911 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | | | |
3912 |{ | | | | |
3913 | top_left_mb[ i ] |1 |ue(v) | | |
3914 | bottom_right_mb[ i ] |1 |ue(v) | | |
3915 | } | | | | |
3916 #endif | |
3917 break; | |
3918 case 3: | |
3919 case 4: | |
3920 case 5: | |
3921 #if 0 | |
3922 | slice_group_change_direction_flag |1 |u(1) | | |
3923 | slice_group_change_rate_minus1 |1 |ue(v) | | |
3924 #endif | |
3925 break; | |
3926 case 6: | |
3927 #if 0 | |
3928 | slice_group_id_cnt_minus1 |1 |ue(v) | | |
3929 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | | | |
3930 |) | | | | |
3931 | slice_group_id[ i ] |1 |u(v) | | |
3932 #endif | |
1214 | 3933 break; |
1168 | 3934 } |
3935 } | |
3936 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1; | |
3937 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1; | |
3938 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){ | |
3939 fprintf(stderr, "reference overflow (pps)\n"); | |
3940 return -1; | |
3941 } | |
3942 | |
3943 pps->weighted_pred= get_bits1(&s->gb); | |
3944 pps->weighted_bipred_idc= get_bits(&s->gb, 2); | |
3945 pps->init_qp= get_se_golomb(&s->gb) + 26; | |
3946 pps->init_qs= get_se_golomb(&s->gb) + 26; | |
3947 pps->chroma_qp_index_offset= get_se_golomb(&s->gb); | |
3948 pps->deblocking_filter_parameters_present= get_bits1(&s->gb); | |
3949 pps->constrained_intra_pred= get_bits1(&s->gb); | |
3950 pps->redundant_pic_cnt_present = get_bits1(&s->gb); | |
3951 pps->crop= get_bits1(&s->gb); | |
3952 if(pps->crop){ | |
3953 pps->crop_left = get_ue_golomb(&s->gb); | |
3954 pps->crop_right = get_ue_golomb(&s->gb); | |
3955 pps->crop_top = get_ue_golomb(&s->gb); | |
3956 pps->crop_bottom= get_ue_golomb(&s->gb); | |
3957 }else{ | |
3958 pps->crop_left = | |
3959 pps->crop_right = | |
3960 pps->crop_top = | |
3961 pps->crop_bottom= 0; | |
3962 } | |
3963 | |
3964 if(s->avctx->debug&FF_DEBUG_PICT_INFO){ | |
3965 printf("pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s crop:%d/%d/%d/%d\n", | |
3966 pps_id, pps->sps_id, | |
3967 pps->cabac ? "CABAC" : "CAVLC", | |
3968 pps->slice_group_count, | |
3969 pps->ref_count[0], pps->ref_count[1], | |
3970 pps->weighted_pred ? "weighted" : "", | |
3971 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset, | |
3972 pps->deblocking_filter_parameters_present ? "LPAR" : "", | |
3973 pps->constrained_intra_pred ? "CONSTR" : "", | |
3974 pps->redundant_pic_cnt_present ? "REDU" : "", | |
3975 pps->crop_left, pps->crop_right, | |
3976 pps->crop_top, pps->crop_bottom | |
3977 ); | |
3978 } | |
3979 | |
3980 return 0; | |
3981 } | |
3982 | |
3983 /** | |
3984 * finds the end of the current frame in the bitstream. | |
3985 * @return the position of the first byte of the next frame, or -1 | |
3986 */ | |
3987 static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){ | |
3988 ParseContext *pc= &s->parse_context; | |
1187 | 3989 int i; |
1168 | 3990 uint32_t state; |
3991 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]); | |
3992 // mb_addr= pc->mb_addr - 1; | |
3993 state= pc->state; | |
3994 //FIXME this will fail with slices | |
3995 for(i=0; i<buf_size; i++){ | |
3996 state= (state<<8) | buf[i]; | |
3997 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){ | |
3998 if(pc->frame_start_found){ | |
3999 pc->state=-1; | |
4000 pc->frame_start_found= 0; | |
4001 return i-3; | |
4002 } | |
4003 pc->frame_start_found= 1; | |
4004 } | |
4005 } | |
4006 | |
4007 pc->state= state; | |
1219 | 4008 return END_NOT_FOUND; |
1168 | 4009 } |
4010 | |
4011 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){ | |
4012 MpegEncContext * const s = &h->s; | |
4013 AVCodecContext * const avctx= s->avctx; | |
4014 int buf_index=0; | |
1322 | 4015 #if 0 |
1168 | 4016 int i; |
4017 for(i=0; i<32; i++){ | |
4018 printf("%X ", buf[i]); | |
4019 } | |
4020 #endif | |
4021 for(;;){ | |
4022 int consumed; | |
4023 int dst_length; | |
4024 int bit_length; | |
4025 uint8_t *ptr; | |
4026 | |
4027 // start code prefix search | |
4028 for(; buf_index + 3 < buf_size; buf_index++){ | |
4029 // this should allways succeed in the first iteration | |
4030 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1) | |
4031 break; | |
4032 } | |
4033 | |
4034 if(buf_index+3 >= buf_size) break; | |
4035 | |
4036 buf_index+=3; | |
4037 | |
4038 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index); | |
4039 if(ptr[dst_length - 1] == 0) dst_length--; | |
4040 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1); | |
4041 | |
4042 if(s->avctx->debug&FF_DEBUG_STARTCODE){ | |
4043 printf("NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length); | |
4044 } | |
4045 | |
4046 buf_index += consumed; | |
4047 | |
4048 if(h->nal_ref_idc < s->hurry_up) | |
4049 continue; | |
4050 | |
4051 switch(h->nal_unit_type){ | |
4052 case NAL_IDR_SLICE: | |
4053 idr(h); //FIXME ensure we dont loose some frames if there is reordering | |
4054 case NAL_SLICE: | |
4055 init_get_bits(&s->gb, ptr, bit_length); | |
4056 h->intra_gb_ptr= | |
4057 h->inter_gb_ptr= &s->gb; | |
4058 s->data_partitioning = 0; | |
4059 | |
4060 if(decode_slice_header(h) < 0) return -1; | |
4061 if(h->redundant_pic_count==0) | |
4062 decode_slice(h); | |
4063 break; | |
4064 case NAL_DPA: | |
4065 init_get_bits(&s->gb, ptr, bit_length); | |
4066 h->intra_gb_ptr= | |
4067 h->inter_gb_ptr= NULL; | |
4068 s->data_partitioning = 1; | |
4069 | |
4070 if(decode_slice_header(h) < 0) return -1; | |
4071 break; | |
4072 case NAL_DPB: | |
4073 init_get_bits(&h->intra_gb, ptr, bit_length); | |
4074 h->intra_gb_ptr= &h->intra_gb; | |
4075 break; | |
4076 case NAL_DPC: | |
4077 init_get_bits(&h->inter_gb, ptr, bit_length); | |
4078 h->inter_gb_ptr= &h->inter_gb; | |
1174 | 4079 |
4080 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning) | |
1168 | 4081 decode_slice(h); |
4082 break; | |
4083 case NAL_SEI: | |
4084 break; | |
4085 case NAL_SPS: | |
4086 init_get_bits(&s->gb, ptr, bit_length); | |
4087 decode_seq_parameter_set(h); | |
4088 | |
4089 if(s->flags& CODEC_FLAG_LOW_DELAY) | |
4090 s->low_delay=1; | |
4091 | |
4092 avctx->has_b_frames= !s->low_delay; | |
4093 break; | |
4094 case NAL_PPS: | |
4095 init_get_bits(&s->gb, ptr, bit_length); | |
4096 | |
4097 decode_picture_parameter_set(h); | |
4098 | |
4099 break; | |
4100 case NAL_PICTURE_DELIMITER: | |
4101 break; | |
4102 case NAL_FILTER_DATA: | |
4103 break; | |
4104 } | |
4105 | |
4106 //FIXME move after where irt is set | |
4107 s->current_picture.pict_type= s->pict_type; | |
4108 s->current_picture.key_frame= s->pict_type == I_TYPE; | |
4109 } | |
4110 | |
1174 | 4111 if(!s->current_picture_ptr) return buf_index; //no frame |
4112 | |
1168 | 4113 h->prev_frame_num_offset= h->frame_num_offset; |
4114 h->prev_frame_num= h->frame_num; | |
4115 if(s->current_picture_ptr->reference){ | |
4116 h->prev_poc_msb= h->poc_msb; | |
4117 h->prev_poc_lsb= h->poc_lsb; | |
4118 } | |
4119 if(s->current_picture_ptr->reference) | |
4120 execute_ref_pic_marking(h, h->mmco, h->mmco_index); | |
4121 else | |
1169 | 4122 assert(h->mmco_index==0); |
1168 | 4123 |
4124 ff_er_frame_end(s); | |
4125 MPV_frame_end(s); | |
4126 | |
4127 return buf_index; | |
4128 } | |
4129 | |
4130 /** | |
4131 * retunrs the number of bytes consumed for building the current frame | |
4132 */ | |
4133 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){ | |
4134 if(s->flags&CODEC_FLAG_TRUNCATED){ | |
4135 pos -= s->parse_context.last_index; | |
4136 if(pos<0) pos=0; // FIXME remove (uneeded?) | |
4137 | |
4138 return pos; | |
4139 }else{ | |
4140 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...) | |
4141 if(pos+10>buf_size) pos=buf_size; // oops ;) | |
4142 | |
4143 return pos; | |
4144 } | |
4145 } | |
4146 | |
4147 static int decode_frame(AVCodecContext *avctx, | |
4148 void *data, int *data_size, | |
4149 uint8_t *buf, int buf_size) | |
4150 { | |
4151 H264Context *h = avctx->priv_data; | |
4152 MpegEncContext *s = &h->s; | |
4153 AVFrame *pict = data; | |
4154 int buf_index; | |
4155 | |
4156 s->flags= avctx->flags; | |
4157 | |
4158 *data_size = 0; | |
4159 | |
4160 /* no supplementary picture */ | |
4161 if (buf_size == 0) { | |
4162 return 0; | |
4163 } | |
4164 | |
4165 if(s->flags&CODEC_FLAG_TRUNCATED){ | |
4166 int next= find_frame_end(s, buf, buf_size); | |
4167 | |
4168 if( ff_combine_frame(s, next, &buf, &buf_size) < 0 ) | |
4169 return buf_size; | |
4170 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index); | |
4171 } | |
4172 | |
4173 if(s->avctx->extradata_size && s->picture_number==0){ | |
4174 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) ) | |
4175 return -1; | |
4176 } | |
4177 | |
4178 buf_index=decode_nal_units(h, buf, buf_size); | |
4179 if(buf_index < 0) | |
4180 return -1; | |
4181 | |
4182 //FIXME do something with unavailable reference frames | |
4183 | |
4184 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size); | |
4185 #if 0 | |
4186 if(s->pict_type==B_TYPE || s->low_delay){ | |
4187 *pict= *(AVFrame*)&s->current_picture; | |
4188 } else { | |
4189 *pict= *(AVFrame*)&s->last_picture; | |
4190 } | |
4191 #endif | |
1174 | 4192 if(!s->current_picture_ptr){ |
4193 fprintf(stderr, "error, NO frame\n"); | |
4194 return -1; | |
4195 } | |
4196 | |
1168 | 4197 *pict= *(AVFrame*)&s->current_picture; //FIXME |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
4198 ff_print_debug_info(s, s->current_picture_ptr); |
1168 | 4199 assert(pict->data[0]); |
4200 //printf("out %d\n", (int)pict->data[0]); | |
4201 #if 0 //? | |
4202 | |
4203 /* Return the Picture timestamp as the frame number */ | |
4204 /* we substract 1 because it is added on utils.c */ | |
4205 avctx->frame_number = s->picture_number - 1; | |
4206 #endif | |
4207 #if 0 | |
4208 /* dont output the last pic after seeking */ | |
4209 if(s->last_picture_ptr || s->low_delay) | |
4210 //Note this isnt a issue as a IDR pic should flush teh buffers | |
4211 #endif | |
4212 *data_size = sizeof(AVFrame); | |
4213 return get_consumed_bytes(s, buf_index, buf_size); | |
4214 } | |
4215 #if 0 | |
4216 static inline void fill_mb_avail(H264Context *h){ | |
4217 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
4218 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 4219 |
4220 if(s->mb_y){ | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
4221 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
4222 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
4223 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num; |
1168 | 4224 }else{ |
4225 h->mb_avail[0]= | |
4226 h->mb_avail[1]= | |
4227 h->mb_avail[2]= 0; | |
4228 } | |
4229 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num; | |
4230 h->mb_avail[4]= 1; //FIXME move out | |
4231 h->mb_avail[5]= 0; //FIXME move out | |
4232 } | |
4233 #endif | |
4234 | |
4235 #if 0 //selftest | |
4236 #define COUNT 8000 | |
4237 #define SIZE (COUNT*40) | |
4238 int main(){ | |
4239 int i; | |
4240 uint8_t temp[SIZE]; | |
4241 PutBitContext pb; | |
4242 GetBitContext gb; | |
4243 // int int_temp[10000]; | |
4244 DSPContext dsp; | |
4245 AVCodecContext avctx; | |
4246 | |
4247 dsputil_init(&dsp, &avctx); | |
4248 | |
4249 init_put_bits(&pb, temp, SIZE, NULL, NULL); | |
4250 printf("testing unsigned exp golomb\n"); | |
4251 for(i=0; i<COUNT; i++){ | |
4252 START_TIMER | |
4253 set_ue_golomb(&pb, i); | |
4254 STOP_TIMER("set_ue_golomb"); | |
4255 } | |
4256 flush_put_bits(&pb); | |
4257 | |
4258 init_get_bits(&gb, temp, 8*SIZE); | |
4259 for(i=0; i<COUNT; i++){ | |
4260 int j, s; | |
4261 | |
4262 s= show_bits(&gb, 24); | |
4263 | |
4264 START_TIMER | |
4265 j= get_ue_golomb(&gb); | |
4266 if(j != i){ | |
4267 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s); | |
4268 // return -1; | |
4269 } | |
4270 STOP_TIMER("get_ue_golomb"); | |
4271 } | |
4272 | |
4273 | |
4274 init_put_bits(&pb, temp, SIZE, NULL, NULL); | |
4275 printf("testing signed exp golomb\n"); | |
4276 for(i=0; i<COUNT; i++){ | |
4277 START_TIMER | |
4278 set_se_golomb(&pb, i - COUNT/2); | |
4279 STOP_TIMER("set_se_golomb"); | |
4280 } | |
4281 flush_put_bits(&pb); | |
4282 | |
4283 init_get_bits(&gb, temp, 8*SIZE); | |
4284 for(i=0; i<COUNT; i++){ | |
4285 int j, s; | |
4286 | |
4287 s= show_bits(&gb, 24); | |
4288 | |
4289 START_TIMER | |
4290 j= get_se_golomb(&gb); | |
4291 if(j != i - COUNT/2){ | |
4292 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s); | |
4293 // return -1; | |
4294 } | |
4295 STOP_TIMER("get_se_golomb"); | |
4296 } | |
4297 | |
4298 printf("testing 4x4 (I)DCT\n"); | |
4299 | |
4300 DCTELEM block[16]; | |
4301 uint8_t src[16], ref[16]; | |
4302 uint64_t error= 0, max_error=0; | |
4303 | |
4304 for(i=0; i<COUNT; i++){ | |
4305 int j; | |
4306 // printf("%d %d %d\n", r1, r2, (r2-r1)*16); | |
4307 for(j=0; j<16; j++){ | |
4308 ref[j]= random()%255; | |
4309 src[j]= random()%255; | |
4310 } | |
4311 | |
4312 h264_diff_dct_c(block, src, ref, 4); | |
4313 | |
4314 //normalize | |
4315 for(j=0; j<16; j++){ | |
4316 // printf("%d ", block[j]); | |
4317 block[j]= block[j]*4; | |
4318 if(j&1) block[j]= (block[j]*4 + 2)/5; | |
4319 if(j&4) block[j]= (block[j]*4 + 2)/5; | |
4320 } | |
4321 // printf("\n"); | |
4322 | |
4323 h264_add_idct_c(ref, block, 4); | |
4324 /* for(j=0; j<16; j++){ | |
4325 printf("%d ", ref[j]); | |
4326 } | |
4327 printf("\n");*/ | |
4328 | |
4329 for(j=0; j<16; j++){ | |
4330 int diff= ABS(src[j] - ref[j]); | |
4331 | |
4332 error+= diff*diff; | |
4333 max_error= FFMAX(max_error, diff); | |
4334 } | |
4335 } | |
4336 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error ); | |
4337 #if 0 | |
4338 printf("testing quantizer\n"); | |
4339 for(qp=0; qp<52; qp++){ | |
4340 for(i=0; i<16; i++) | |
4341 src1_block[i]= src2_block[i]= random()%255; | |
4342 | |
4343 } | |
4344 #endif | |
4345 printf("Testing NAL layer\n"); | |
4346 | |
4347 uint8_t bitstream[COUNT]; | |
4348 uint8_t nal[COUNT*2]; | |
4349 H264Context h; | |
4350 memset(&h, 0, sizeof(H264Context)); | |
4351 | |
4352 for(i=0; i<COUNT; i++){ | |
4353 int zeros= i; | |
4354 int nal_length; | |
4355 int consumed; | |
4356 int out_length; | |
4357 uint8_t *out; | |
4358 int j; | |
4359 | |
4360 for(j=0; j<COUNT; j++){ | |
4361 bitstream[j]= (random() % 255) + 1; | |
4362 } | |
4363 | |
4364 for(j=0; j<zeros; j++){ | |
4365 int pos= random() % COUNT; | |
4366 while(bitstream[pos] == 0){ | |
4367 pos++; | |
4368 pos %= COUNT; | |
4369 } | |
4370 bitstream[pos]=0; | |
4371 } | |
4372 | |
4373 START_TIMER | |
4374 | |
4375 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2); | |
4376 if(nal_length<0){ | |
4377 printf("encoding failed\n"); | |
4378 return -1; | |
4379 } | |
4380 | |
4381 out= decode_nal(&h, nal, &out_length, &consumed, nal_length); | |
4382 | |
4383 STOP_TIMER("NAL") | |
4384 | |
4385 if(out_length != COUNT){ | |
4386 printf("incorrect length %d %d\n", out_length, COUNT); | |
4387 return -1; | |
4388 } | |
4389 | |
4390 if(consumed != nal_length){ | |
4391 printf("incorrect consumed length %d %d\n", nal_length, consumed); | |
4392 return -1; | |
4393 } | |
4394 | |
4395 if(memcmp(bitstream, out, COUNT)){ | |
4396 printf("missmatch\n"); | |
4397 return -1; | |
4398 } | |
4399 } | |
4400 | |
4401 printf("Testing RBSP\n"); | |
4402 | |
4403 | |
4404 return 0; | |
4405 } | |
4406 #endif | |
4407 | |
4408 | |
4409 static int decode_end(AVCodecContext *avctx) | |
4410 { | |
4411 H264Context *h = avctx->priv_data; | |
4412 MpegEncContext *s = &h->s; | |
4413 | |
4414 free_tables(h); //FIXME cleanup init stuff perhaps | |
4415 MPV_common_end(s); | |
4416 | |
4417 // memset(h, 0, sizeof(H264Context)); | |
4418 | |
4419 return 0; | |
4420 } | |
4421 | |
4422 | |
4423 AVCodec h264_decoder = { | |
4424 "h264", | |
4425 CODEC_TYPE_VIDEO, | |
4426 CODEC_ID_H264, | |
4427 sizeof(H264Context), | |
4428 decode_init, | |
4429 NULL, | |
4430 decode_end, | |
4431 decode_frame, | |
1214 | 4432 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED, |
1168 | 4433 }; |
4434 | |
1234 | 4435 #include "svq3.c" |