Mercurial > libavcodec.hg
annotate h264.c @ 1898:7d2907127da3 libavcodec
h264 loop filter optimizations
author | michael |
---|---|
date | Sat, 20 Mar 2004 00:18:52 +0000 |
parents | 4e8ed93524f6 |
children | 379a18a7131f |
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 log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4 | |
63 int poc_type; ///< pic_order_cnt_type | |
64 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4 | |
65 int delta_pic_order_always_zero_flag; | |
66 int offset_for_non_ref_pic; | |
67 int offset_for_top_to_bottom_field; | |
68 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle | |
69 int ref_frame_count; ///< num_ref_frames | |
1371 | 70 int gaps_in_frame_num_allowed_flag; |
1168 | 71 int mb_width; ///< frame_width_in_mbs_minus1 + 1 |
72 int mb_height; ///< frame_height_in_mbs_minus1 + 1 | |
73 int frame_mbs_only_flag; | |
74 int mb_aff; ///<mb_adaptive_frame_field_flag | |
75 int direct_8x8_inference_flag; | |
1371 | 76 int crop; ///< frame_cropping_flag |
77 int crop_left; ///< frame_cropping_rect_left_offset | |
78 int crop_right; ///< frame_cropping_rect_right_offset | |
79 int crop_top; ///< frame_cropping_rect_top_offset | |
80 int crop_bottom; ///< frame_cropping_rect_bottom_offset | |
1168 | 81 int vui_parameters_present_flag; |
1548 | 82 AVRational sar; |
1168 | 83 short offset_for_ref_frame[256]; //FIXME dyn aloc? |
84 }SPS; | |
85 | |
86 /** | |
87 * Picture parameter set | |
88 */ | |
89 typedef struct PPS{ | |
90 int sps_id; | |
91 int cabac; ///< entropy_coding_mode_flag | |
92 int pic_order_present; ///< pic_order_present_flag | |
93 int slice_group_count; ///< num_slice_groups_minus1 + 1 | |
94 int mb_slice_group_map_type; | |
95 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1 | |
96 int weighted_pred; ///< weighted_pred_flag | |
97 int weighted_bipred_idc; | |
98 int init_qp; ///< pic_init_qp_minus26 + 26 | |
99 int init_qs; ///< pic_init_qs_minus26 + 26 | |
100 int chroma_qp_index_offset; | |
101 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag | |
102 int constrained_intra_pred; ///< constrained_intra_pred_flag | |
103 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag | |
104 }PPS; | |
105 | |
106 /** | |
107 * Memory management control operation opcode. | |
108 */ | |
109 typedef enum MMCOOpcode{ | |
110 MMCO_END=0, | |
111 MMCO_SHORT2UNUSED, | |
112 MMCO_LONG2UNUSED, | |
113 MMCO_SHORT2LONG, | |
114 MMCO_SET_MAX_LONG, | |
115 MMCO_RESET, | |
116 MMCO_LONG, | |
117 } MMCOOpcode; | |
118 | |
119 /** | |
120 * Memory management control operation. | |
121 */ | |
122 typedef struct MMCO{ | |
123 MMCOOpcode opcode; | |
124 int short_frame_num; | |
125 int long_index; | |
126 } MMCO; | |
127 | |
128 /** | |
129 * H264Context | |
130 */ | |
131 typedef struct H264Context{ | |
132 MpegEncContext s; | |
133 int nal_ref_idc; | |
134 int nal_unit_type; | |
135 #define NAL_SLICE 1 | |
136 #define NAL_DPA 2 | |
137 #define NAL_DPB 3 | |
138 #define NAL_DPC 4 | |
139 #define NAL_IDR_SLICE 5 | |
140 #define NAL_SEI 6 | |
141 #define NAL_SPS 7 | |
142 #define NAL_PPS 8 | |
143 #define NAL_PICTURE_DELIMITER 9 | |
144 #define NAL_FILTER_DATA 10 | |
145 uint8_t *rbsp_buffer; | |
146 int rbsp_buffer_size; | |
147 | |
148 int chroma_qp; //QPc | |
149 | |
150 int prev_mb_skiped; //FIXME remove (IMHO not used) | |
151 | |
152 //prediction stuff | |
153 int chroma_pred_mode; | |
154 int intra16x16_pred_mode; | |
155 | |
156 int8_t intra4x4_pred_mode_cache[5*8]; | |
157 int8_t (*intra4x4_pred_mode)[8]; | |
158 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp? | |
159 void (*pred8x8 [4+3])(uint8_t *src, int stride); | |
160 void (*pred16x16[4+3])(uint8_t *src, int stride); | |
161 unsigned int topleft_samples_available; | |
162 unsigned int top_samples_available; | |
163 unsigned int topright_samples_available; | |
164 unsigned int left_samples_available; | |
165 | |
166 /** | |
167 * non zero coeff count cache. | |
168 * is 64 if not available. | |
169 */ | |
170 uint8_t non_zero_count_cache[6*8]; | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
171 uint8_t (*non_zero_count)[16+4+4]; /* store all values for deblocking filter */ |
1168 | 172 |
173 /** | |
174 * Motion vector cache. | |
175 */ | |
176 int16_t mv_cache[2][5*8][2]; | |
177 int8_t ref_cache[2][5*8]; | |
178 #define LIST_NOT_USED -1 //FIXME rename? | |
179 #define PART_NOT_AVAILABLE -2 | |
180 | |
181 /** | |
182 * is 1 if the specific list MV&references are set to 0,0,-2. | |
183 */ | |
184 int mv_cache_clean[2]; | |
185 | |
186 int block_offset[16+8]; | |
187 int chroma_subblock_offset[16]; //FIXME remove | |
188 | |
189 uint16_t *mb2b_xy; //FIXME are these 4 a good idea? | |
190 uint16_t *mb2b8_xy; | |
191 int b_stride; | |
192 int b8_stride; | |
193 | |
1234 | 194 int halfpel_flag; |
195 int thirdpel_flag; | |
196 | |
1319 | 197 int unknown_svq3_flag; |
198 int next_slice_index; | |
199 | |
1168 | 200 SPS sps_buffer[MAX_SPS_COUNT]; |
201 SPS sps; ///< current sps | |
202 | |
203 PPS pps_buffer[MAX_PPS_COUNT]; | |
204 /** | |
205 * current pps | |
206 */ | |
207 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that? | |
208 | |
209 int slice_num; | |
210 uint8_t *slice_table_base; | |
211 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1 | |
212 int slice_type; | |
213 int slice_type_fixed; | |
214 | |
215 //interlacing specific flags | |
216 int mb_field_decoding_flag; | |
217 | |
218 int sub_mb_type[4]; | |
219 | |
220 //POC stuff | |
221 int poc_lsb; | |
222 int poc_msb; | |
223 int delta_poc_bottom; | |
224 int delta_poc[2]; | |
225 int frame_num; | |
226 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0 | |
227 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0 | |
228 int frame_num_offset; ///< for POC type 2 | |
229 int prev_frame_num_offset; ///< for POC type 2 | |
230 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2 | |
231 | |
232 /** | |
233 * frame_num for frames or 2*frame_num for field pics. | |
234 */ | |
235 int curr_pic_num; | |
236 | |
237 /** | |
238 * max_frame_num or 2*max_frame_num for field pics. | |
239 */ | |
240 int max_pic_num; | |
241 | |
242 //Weighted pred stuff | |
243 int luma_log2_weight_denom; | |
244 int chroma_log2_weight_denom; | |
245 int luma_weight[2][16]; | |
246 int luma_offset[2][16]; | |
247 int chroma_weight[2][16][2]; | |
248 int chroma_offset[2][16][2]; | |
249 | |
250 //deblock | |
251 int disable_deblocking_filter_idc; | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
252 int slice_alpha_c0_offset; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
253 int slice_beta_offset; |
1168 | 254 |
255 int redundant_pic_count; | |
256 | |
257 int direct_spatial_mv_pred; | |
258 | |
259 /** | |
260 * num_ref_idx_l0/1_active_minus1 + 1 | |
261 */ | |
262 int ref_count[2];// FIXME split for AFF | |
263 Picture *short_ref[16]; | |
264 Picture *long_ref[16]; | |
265 Picture default_ref_list[2][32]; | |
266 Picture ref_list[2][32]; //FIXME size? | |
267 Picture field_ref_list[2][32]; //FIXME size? | |
268 | |
269 /** | |
270 * memory management control operations buffer. | |
271 */ | |
272 MMCO mmco[MAX_MMCO_COUNT]; | |
273 int mmco_index; | |
274 | |
275 int long_ref_count; ///< number of actual long term references | |
276 int short_ref_count; ///< number of actual short term references | |
277 | |
278 //data partitioning | |
279 GetBitContext intra_gb; | |
280 GetBitContext inter_gb; | |
281 GetBitContext *intra_gb_ptr; | |
282 GetBitContext *inter_gb_ptr; | |
283 | |
284 DCTELEM mb[16*24] __align8; | |
285 }H264Context; | |
286 | |
287 static VLC coeff_token_vlc[4]; | |
288 static VLC chroma_dc_coeff_token_vlc; | |
289 | |
290 static VLC total_zeros_vlc[15]; | |
291 static VLC chroma_dc_total_zeros_vlc[3]; | |
292 | |
293 static VLC run_vlc[6]; | |
294 static VLC run7_vlc; | |
295 | |
1234 | 296 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp); |
297 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc); | |
298 | |
1269 | 299 static inline uint32_t pack16to32(int a, int b){ |
300 #ifdef WORDS_BIGENDIAN | |
301 return (b&0xFFFF) + (a<<16); | |
302 #else | |
303 return (a&0xFFFF) + (b<<16); | |
304 #endif | |
305 } | |
306 | |
1168 | 307 /** |
308 * fill a rectangle. | |
309 * @param h height of the recatangle, should be a constant | |
310 * @param w width of the recatangle, should be a constant | |
311 * @param size the size of val (1 or 4), should be a constant | |
312 */ | |
1187 | 313 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined |
314 uint8_t *p= (uint8_t*)vp; | |
1168 | 315 assert(size==1 || size==4); |
316 | |
317 w *= size; | |
318 stride *= size; | |
319 | |
320 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it | |
321 if(w==2 && h==2){ | |
322 *(uint16_t*)(p + 0)= | |
323 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101; | |
324 }else if(w==2 && h==4){ | |
325 *(uint16_t*)(p + 0*stride)= | |
326 *(uint16_t*)(p + 1*stride)= | |
327 *(uint16_t*)(p + 2*stride)= | |
328 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101; | |
1252 | 329 }else if(w==4 && h==1){ |
330 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101; | |
1168 | 331 }else if(w==4 && h==2){ |
332 *(uint32_t*)(p + 0*stride)= | |
333 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101; | |
334 }else if(w==4 && h==4){ | |
335 *(uint32_t*)(p + 0*stride)= | |
336 *(uint32_t*)(p + 1*stride)= | |
337 *(uint32_t*)(p + 2*stride)= | |
338 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101; | |
339 }else if(w==8 && h==1){ | |
340 *(uint32_t*)(p + 0)= | |
341 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101; | |
342 }else if(w==8 && h==2){ | |
343 *(uint32_t*)(p + 0 + 0*stride)= | |
344 *(uint32_t*)(p + 4 + 0*stride)= | |
345 *(uint32_t*)(p + 0 + 1*stride)= | |
346 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101; | |
347 }else if(w==8 && h==4){ | |
348 *(uint64_t*)(p + 0*stride)= | |
349 *(uint64_t*)(p + 1*stride)= | |
350 *(uint64_t*)(p + 2*stride)= | |
351 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL; | |
352 }else if(w==16 && h==2){ | |
353 *(uint64_t*)(p + 0+0*stride)= | |
354 *(uint64_t*)(p + 8+0*stride)= | |
355 *(uint64_t*)(p + 0+1*stride)= | |
356 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL; | |
357 }else if(w==16 && h==4){ | |
358 *(uint64_t*)(p + 0+0*stride)= | |
359 *(uint64_t*)(p + 8+0*stride)= | |
360 *(uint64_t*)(p + 0+1*stride)= | |
361 *(uint64_t*)(p + 8+1*stride)= | |
362 *(uint64_t*)(p + 0+2*stride)= | |
363 *(uint64_t*)(p + 8+2*stride)= | |
364 *(uint64_t*)(p + 0+3*stride)= | |
365 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL; | |
366 }else | |
367 assert(0); | |
368 } | |
369 | |
370 static inline void fill_caches(H264Context *h, int mb_type){ | |
371 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
|
372 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 373 int topleft_xy, top_xy, topright_xy, left_xy[2]; |
374 int topleft_type, top_type, topright_type, left_type[2]; | |
375 int left_block[4]; | |
376 int i; | |
377 | |
378 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it | |
379 | |
380 if(h->sps.mb_aff){ | |
381 //FIXME | |
1453 | 382 topleft_xy = 0; /* avoid warning */ |
383 top_xy = 0; /* avoid warning */ | |
384 topright_xy = 0; /* avoid warning */ | |
1168 | 385 }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
|
386 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
|
387 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
|
388 topright_xy= mb_xy+1 - s->mb_stride; |
1168 | 389 left_xy[0] = mb_xy-1; |
390 left_xy[1] = mb_xy-1; | |
391 left_block[0]= 0; | |
392 left_block[1]= 1; | |
393 left_block[2]= 2; | |
394 left_block[3]= 3; | |
395 } | |
396 | |
397 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0; | |
398 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0; | |
399 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0; | |
400 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0; | |
401 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0; | |
402 | |
403 if(IS_INTRA(mb_type)){ | |
404 h->topleft_samples_available= | |
405 h->top_samples_available= | |
406 h->left_samples_available= 0xFFFF; | |
407 h->topright_samples_available= 0xEEEA; | |
408 | |
409 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){ | |
410 h->topleft_samples_available= 0xB3FF; | |
411 h->top_samples_available= 0x33FF; | |
412 h->topright_samples_available= 0x26EA; | |
413 } | |
414 for(i=0; i<2; i++){ | |
415 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){ | |
416 h->topleft_samples_available&= 0xDF5F; | |
417 h->left_samples_available&= 0x5F5F; | |
418 } | |
419 } | |
420 | |
421 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred)) | |
422 h->topleft_samples_available&= 0x7FFF; | |
423 | |
424 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred)) | |
425 h->topright_samples_available&= 0xFBFF; | |
426 | |
427 if(IS_INTRA4x4(mb_type)){ | |
428 if(IS_INTRA4x4(top_type)){ | |
429 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4]; | |
430 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5]; | |
431 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6]; | |
432 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3]; | |
433 }else{ | |
434 int pred; | |
435 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred)) | |
436 pred= 2; | |
437 else{ | |
438 pred= -1; | |
439 } | |
440 h->intra4x4_pred_mode_cache[4+8*0]= | |
441 h->intra4x4_pred_mode_cache[5+8*0]= | |
442 h->intra4x4_pred_mode_cache[6+8*0]= | |
443 h->intra4x4_pred_mode_cache[7+8*0]= pred; | |
444 } | |
445 for(i=0; i<2; i++){ | |
446 if(IS_INTRA4x4(left_type[i])){ | |
447 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]]; | |
448 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]]; | |
449 }else{ | |
450 int pred; | |
451 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred)) | |
452 pred= 2; | |
453 else{ | |
454 pred= -1; | |
455 } | |
456 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= | |
457 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred; | |
458 } | |
459 } | |
460 } | |
461 } | |
462 | |
463 | |
464 /* | |
465 0 . T T. T T T T | |
466 1 L . .L . . . . | |
467 2 L . .L . . . . | |
468 3 . T TL . . . . | |
469 4 L . .L . . . . | |
470 5 L . .. . . . . | |
471 */ | |
472 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec) | |
473 if(top_type){ | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
474 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][10]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
475 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][11]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
476 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][14]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
477 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][15]; |
1168 | 478 |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
479 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][18]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
480 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][19]; |
1168 | 481 |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
482 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][22]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
483 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][23]; |
1168 | 484 }else{ |
485 h->non_zero_count_cache[4+8*0]= | |
486 h->non_zero_count_cache[5+8*0]= | |
487 h->non_zero_count_cache[6+8*0]= | |
488 h->non_zero_count_cache[7+8*0]= | |
489 | |
490 h->non_zero_count_cache[1+8*0]= | |
491 h->non_zero_count_cache[2+8*0]= | |
492 | |
493 h->non_zero_count_cache[1+8*3]= | |
494 h->non_zero_count_cache[2+8*3]= 64; | |
495 } | |
496 | |
497 if(left_type[0]){ | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
498 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][5]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
499 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
500 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][17]; //FIXME left_block |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
501 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][21]; |
1168 | 502 }else{ |
503 h->non_zero_count_cache[3+8*1]= | |
504 h->non_zero_count_cache[3+8*2]= | |
505 h->non_zero_count_cache[0+8*1]= | |
506 h->non_zero_count_cache[0+8*4]= 64; | |
507 } | |
508 | |
509 if(left_type[1]){ | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
510 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][13]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
511 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][15]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
512 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][19]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
513 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][23]; |
1168 | 514 }else{ |
515 h->non_zero_count_cache[3+8*3]= | |
516 h->non_zero_count_cache[3+8*4]= | |
517 h->non_zero_count_cache[0+8*2]= | |
518 h->non_zero_count_cache[0+8*5]= 64; | |
519 } | |
520 | |
521 #if 1 | |
522 if(IS_INTER(mb_type)){ | |
523 int list; | |
524 for(list=0; list<2; list++){ | |
525 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){ | |
526 /*if(!h->mv_cache_clean[list]){ | |
527 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all? | |
528 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t)); | |
529 h->mv_cache_clean[list]= 1; | |
530 }*/ | |
531 continue; //FIXME direct mode ... | |
532 } | |
533 h->mv_cache_clean[list]= 0; | |
534 | |
535 if(IS_INTER(topleft_type)){ | |
536 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride; | |
537 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride; | |
538 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy]; | |
539 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy]; | |
540 }else{ | |
541 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0; | |
542 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
543 } | |
544 | |
545 if(IS_INTER(top_type)){ | |
546 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride; | |
547 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride; | |
548 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0]; | |
549 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1]; | |
550 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2]; | |
551 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3]; | |
552 h->ref_cache[list][scan8[0] + 0 - 1*8]= | |
553 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0]; | |
554 h->ref_cache[list][scan8[0] + 2 - 1*8]= | |
555 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1]; | |
556 }else{ | |
557 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= | |
558 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= | |
559 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= | |
560 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0; | |
561 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101; | |
562 } | |
563 | |
564 if(IS_INTER(topright_type)){ | |
565 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride; | |
566 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride; | |
567 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy]; | |
568 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy]; | |
569 }else{ | |
570 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0; | |
571 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
572 } | |
573 | |
574 //FIXME unify cleanup or sth | |
575 if(IS_INTER(left_type[0])){ | |
576 const int b_xy= h->mb2b_xy[left_xy[0]] + 3; | |
577 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1; | |
578 *(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]]; | |
579 *(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]]; | |
580 h->ref_cache[list][scan8[0] - 1 + 0*8]= | |
581 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)]; | |
582 }else{ | |
583 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]= | |
584 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0; | |
585 h->ref_cache[list][scan8[0] - 1 + 0*8]= | |
586 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
587 } | |
588 | |
589 if(IS_INTER(left_type[1])){ | |
590 const int b_xy= h->mb2b_xy[left_xy[1]] + 3; | |
591 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1; | |
592 *(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]]; | |
593 *(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]]; | |
594 h->ref_cache[list][scan8[0] - 1 + 2*8]= | |
595 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)]; | |
596 }else{ | |
597 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]= | |
598 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0; | |
599 h->ref_cache[list][scan8[0] - 1 + 2*8]= | |
600 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
601 } | |
602 | |
603 h->ref_cache[list][scan8[5 ]+1] = | |
604 h->ref_cache[list][scan8[7 ]+1] = | |
605 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else) | |
606 h->ref_cache[list][scan8[4 ]] = | |
607 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE; | |
608 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]= | |
609 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]= | |
610 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else) | |
611 *(uint32_t*)h->mv_cache [list][scan8[4 ]]= | |
612 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0; | |
613 } | |
614 //FIXME | |
615 | |
616 } | |
617 #endif | |
618 } | |
619 | |
620 static inline void write_back_intra_pred_mode(H264Context *h){ | |
621 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
|
622 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 623 |
624 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1]; | |
625 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2]; | |
626 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3]; | |
627 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4]; | |
628 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4]; | |
629 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4]; | |
630 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4]; | |
631 } | |
632 | |
633 /** | |
634 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. | |
635 */ | |
636 static inline int check_intra4x4_pred_mode(H264Context *h){ | |
637 MpegEncContext * const s = &h->s; | |
638 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0}; | |
639 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED}; | |
640 int i; | |
641 | |
642 if(!(h->top_samples_available&0x8000)){ | |
643 for(i=0; i<4; i++){ | |
644 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ]; | |
645 if(status<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
646 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y); |
1168 | 647 return -1; |
648 } else if(status){ | |
649 h->intra4x4_pred_mode_cache[scan8[0] + i]= status; | |
650 } | |
651 } | |
652 } | |
653 | |
654 if(!(h->left_samples_available&0x8000)){ | |
655 for(i=0; i<4; i++){ | |
656 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ]; | |
657 if(status<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
658 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y); |
1168 | 659 return -1; |
660 } else if(status){ | |
661 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status; | |
662 } | |
663 } | |
664 } | |
665 | |
666 return 0; | |
667 } //FIXME cleanup like next | |
668 | |
669 /** | |
670 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. | |
671 */ | |
672 static inline int check_intra_pred_mode(H264Context *h, int mode){ | |
673 MpegEncContext * const s = &h->s; | |
674 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1}; | |
675 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8}; | |
676 | |
677 if(!(h->top_samples_available&0x8000)){ | |
678 mode= top[ mode ]; | |
679 if(mode<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
680 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y); |
1168 | 681 return -1; |
682 } | |
683 } | |
684 | |
685 if(!(h->left_samples_available&0x8000)){ | |
686 mode= left[ mode ]; | |
687 if(mode<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
688 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y); |
1168 | 689 return -1; |
690 } | |
691 } | |
692 | |
693 return mode; | |
694 } | |
695 | |
696 /** | |
697 * gets the predicted intra4x4 prediction mode. | |
698 */ | |
699 static inline int pred_intra_mode(H264Context *h, int n){ | |
700 const int index8= scan8[n]; | |
701 const int left= h->intra4x4_pred_mode_cache[index8 - 1]; | |
702 const int top = h->intra4x4_pred_mode_cache[index8 - 8]; | |
703 const int min= FFMIN(left, top); | |
704 | |
1170 | 705 tprintf("mode:%d %d min:%d\n", left ,top, min); |
1168 | 706 |
707 if(min<0) return DC_PRED; | |
708 else return min; | |
709 } | |
710 | |
711 static inline void write_back_non_zero_count(H264Context *h){ | |
712 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
|
713 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
714 int n; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
715 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
716 for( n = 0; n < 16+4+4; n++ ) |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
717 h->non_zero_count[mb_xy][n] = h->non_zero_count_cache[scan8[n]]; |
1168 | 718 } |
719 | |
720 /** | |
721 * gets the predicted number of non zero coefficients. | |
722 * @param n block index | |
723 */ | |
724 static inline int pred_non_zero_count(H264Context *h, int n){ | |
725 const int index8= scan8[n]; | |
726 const int left= h->non_zero_count_cache[index8 - 1]; | |
727 const int top = h->non_zero_count_cache[index8 - 8]; | |
728 int i= left + top; | |
729 | |
730 if(i<64) i= (i+1)>>1; | |
731 | |
1170 | 732 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31); |
1168 | 733 |
734 return i&31; | |
735 } | |
736 | |
1169 | 737 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){ |
738 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ]; | |
739 | |
740 if(topright_ref != PART_NOT_AVAILABLE){ | |
741 *C= h->mv_cache[list][ i - 8 + part_width ]; | |
742 return topright_ref; | |
743 }else{ | |
1170 | 744 tprintf("topright MV not available\n"); |
745 | |
1169 | 746 *C= h->mv_cache[list][ i - 8 - 1 ]; |
747 return h->ref_cache[list][ i - 8 - 1 ]; | |
748 } | |
749 } | |
750 | |
1168 | 751 /** |
752 * gets the predicted MV. | |
753 * @param n the block index | |
754 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4) | |
755 * @param mx the x component of the predicted motion vector | |
756 * @param my the y component of the predicted motion vector | |
757 */ | |
758 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){ | |
759 const int index8= scan8[n]; | |
760 const int top_ref= h->ref_cache[list][ index8 - 8 ]; | |
761 const int left_ref= h->ref_cache[list][ index8 - 1 ]; | |
762 const int16_t * const A= h->mv_cache[list][ index8 - 1 ]; | |
763 const int16_t * const B= h->mv_cache[list][ index8 - 8 ]; | |
1169 | 764 const int16_t * C; |
765 int diagonal_ref, match_count; | |
766 | |
1168 | 767 assert(part_width==1 || part_width==2 || part_width==4); |
1169 | 768 |
1168 | 769 /* mv_cache |
770 B . . A T T T T | |
771 U . . L . . , . | |
772 U . . L . . . . | |
773 U . . L . . , . | |
774 . . . L . . . . | |
775 */ | |
1169 | 776 |
777 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width); | |
778 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref); | |
779 if(match_count > 1){ //most common | |
780 *mx= mid_pred(A[0], B[0], C[0]); | |
781 *my= mid_pred(A[1], B[1], C[1]); | |
782 }else if(match_count==1){ | |
783 if(left_ref==ref){ | |
784 *mx= A[0]; | |
785 *my= A[1]; | |
786 }else if(top_ref==ref){ | |
787 *mx= B[0]; | |
788 *my= B[1]; | |
789 }else{ | |
790 *mx= C[0]; | |
791 *my= C[1]; | |
792 } | |
793 }else{ | |
794 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){ | |
795 *mx= A[0]; | |
796 *my= A[1]; | |
1168 | 797 }else{ |
798 *mx= mid_pred(A[0], B[0], C[0]); | |
799 *my= mid_pred(A[1], B[1], C[1]); | |
800 } | |
1169 | 801 } |
1168 | 802 |
1187 | 803 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 | 804 } |
805 | |
806 /** | |
807 * gets the directionally predicted 16x8 MV. | |
808 * @param n the block index | |
809 * @param mx the x component of the predicted motion vector | |
810 * @param my the y component of the predicted motion vector | |
811 */ | |
812 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){ | |
813 if(n==0){ | |
814 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ]; | |
815 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ]; | |
816 | |
1187 | 817 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 | 818 |
819 if(top_ref == ref){ | |
820 *mx= B[0]; | |
821 *my= B[1]; | |
822 return; | |
823 } | |
824 }else{ | |
825 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ]; | |
826 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ]; | |
827 | |
1187 | 828 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 | 829 |
830 if(left_ref == ref){ | |
831 *mx= A[0]; | |
832 *my= A[1]; | |
833 return; | |
834 } | |
835 } | |
836 | |
837 //RARE | |
838 pred_motion(h, n, 4, list, ref, mx, my); | |
839 } | |
840 | |
841 /** | |
842 * gets the directionally predicted 8x16 MV. | |
843 * @param n the block index | |
844 * @param mx the x component of the predicted motion vector | |
845 * @param my the y component of the predicted motion vector | |
846 */ | |
847 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){ | |
848 if(n==0){ | |
849 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ]; | |
850 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ]; | |
851 | |
1187 | 852 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 | 853 |
854 if(left_ref == ref){ | |
855 *mx= A[0]; | |
856 *my= A[1]; | |
857 return; | |
858 } | |
859 }else{ | |
1169 | 860 const int16_t * C; |
861 int diagonal_ref; | |
862 | |
863 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2); | |
1168 | 864 |
1187 | 865 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 | 866 |
1169 | 867 if(diagonal_ref == ref){ |
1168 | 868 *mx= C[0]; |
869 *my= C[1]; | |
870 return; | |
871 } | |
872 } | |
873 | |
874 //RARE | |
875 pred_motion(h, n, 2, list, ref, mx, my); | |
876 } | |
877 | |
878 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){ | |
879 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ]; | |
880 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ]; | |
881 | |
1187 | 882 tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y); |
1168 | 883 |
884 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE | |
885 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0) | |
886 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){ | |
887 | |
888 *mx = *my = 0; | |
889 return; | |
890 } | |
891 | |
892 pred_motion(h, 0, 4, 0, 0, mx, my); | |
893 | |
894 return; | |
895 } | |
896 | |
897 static inline void write_back_motion(H264Context *h, int mb_type){ | |
898 MpegEncContext * const s = &h->s; | |
899 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; | |
900 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride; | |
901 int list; | |
902 | |
903 for(list=0; list<2; list++){ | |
904 int y; | |
905 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){ | |
906 if(1){ //FIXME skip or never read if mb_type doesnt use it | |
907 for(y=0; y<4; y++){ | |
908 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= | |
909 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0; | |
910 } | |
911 for(y=0; y<2; y++){ | |
912 *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101; | |
913 } | |
914 } | |
915 continue; //FIXME direct mode ... | |
916 } | |
917 | |
918 for(y=0; y<4; y++){ | |
919 *(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]; | |
920 *(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]; | |
921 } | |
922 for(y=0; y<2; y++){ | |
923 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y]; | |
924 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y]; | |
925 } | |
926 } | |
927 } | |
928 | |
929 /** | |
930 * Decodes a network abstraction layer unit. | |
931 * @param consumed is the number of bytes used as input | |
932 * @param length is the length of the array | |
933 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing? | |
934 * @returns decoded bytes, might be src+1 if no escapes | |
935 */ | |
936 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){ | |
937 int i, si, di; | |
938 uint8_t *dst; | |
939 | |
940 // src[0]&0x80; //forbidden bit | |
941 h->nal_ref_idc= src[0]>>5; | |
942 h->nal_unit_type= src[0]&0x1F; | |
943 | |
944 src++; length--; | |
945 #if 0 | |
946 for(i=0; i<length; i++) | |
947 printf("%2X ", src[i]); | |
948 #endif | |
949 for(i=0; i+1<length; i+=2){ | |
950 if(src[i]) continue; | |
951 if(i>0 && src[i-1]==0) i--; | |
952 if(i+2<length && src[i+1]==0 && src[i+2]<=3){ | |
953 if(src[i+2]!=3){ | |
954 /* startcode, so we must be past the end */ | |
955 length=i; | |
956 } | |
957 break; | |
958 } | |
959 } | |
960 | |
961 if(i>=length-1){ //no escaped 0 | |
962 *dst_length= length; | |
963 *consumed= length+1; //+1 for the header | |
964 return src; | |
965 } | |
966 | |
967 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length); | |
968 dst= h->rbsp_buffer; | |
969 | |
970 //printf("deoding esc\n"); | |
971 si=di=0; | |
972 while(si<length){ | |
973 //remove escapes (very rare 1:2^22) | |
974 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){ | |
975 if(src[si+2]==3){ //escape | |
976 dst[di++]= 0; | |
977 dst[di++]= 0; | |
978 si+=3; | |
979 }else //next start code | |
980 break; | |
981 } | |
982 | |
983 dst[di++]= src[si++]; | |
984 } | |
985 | |
986 *dst_length= di; | |
987 *consumed= si + 1;//+1 for the header | |
988 //FIXME store exact number of bits in the getbitcontext (its needed for decoding) | |
989 return dst; | |
990 } | |
991 | |
992 /** | |
993 * @param src the data which should be escaped | |
994 * @param dst the target buffer, dst+1 == src is allowed as a special case | |
995 * @param length the length of the src data | |
996 * @param dst_length the length of the dst array | |
997 * @returns length of escaped data in bytes or -1 if an error occured | |
998 */ | |
999 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){ | |
1000 int i, escape_count, si, di; | |
1001 uint8_t *temp; | |
1002 | |
1003 assert(length>=0); | |
1004 assert(dst_length>0); | |
1005 | |
1006 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type; | |
1007 | |
1008 if(length==0) return 1; | |
1009 | |
1010 escape_count= 0; | |
1011 for(i=0; i<length; i+=2){ | |
1012 if(src[i]) continue; | |
1013 if(i>0 && src[i-1]==0) | |
1014 i--; | |
1015 if(i+2<length && src[i+1]==0 && src[i+2]<=3){ | |
1016 escape_count++; | |
1017 i+=2; | |
1018 } | |
1019 } | |
1020 | |
1021 if(escape_count==0){ | |
1022 if(dst+1 != src) | |
1023 memcpy(dst+1, src, length); | |
1024 return length + 1; | |
1025 } | |
1026 | |
1027 if(length + escape_count + 1> dst_length) | |
1028 return -1; | |
1029 | |
1030 //this should be damn rare (hopefully) | |
1031 | |
1032 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count); | |
1033 temp= h->rbsp_buffer; | |
1034 //printf("encoding esc\n"); | |
1035 | |
1036 si= 0; | |
1037 di= 0; | |
1038 while(si < length){ | |
1039 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){ | |
1040 temp[di++]= 0; si++; | |
1041 temp[di++]= 0; si++; | |
1042 temp[di++]= 3; | |
1043 temp[di++]= src[si++]; | |
1044 } | |
1045 else | |
1046 temp[di++]= src[si++]; | |
1047 } | |
1048 memcpy(dst+1, temp, length+escape_count); | |
1049 | |
1050 assert(di == length+escape_count); | |
1051 | |
1052 return di + 1; | |
1053 } | |
1054 | |
1055 /** | |
1056 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4 | |
1057 */ | |
1058 static void encode_rbsp_trailing(PutBitContext *pb){ | |
1059 int length; | |
1060 put_bits(pb, 1, 1); | |
1786 | 1061 length= (-put_bits_count(pb))&7; |
1168 | 1062 if(length) put_bits(pb, length, 0); |
1063 } | |
1064 | |
1065 /** | |
1066 * identifies the exact end of the bitstream | |
1067 * @return the length of the trailing, or 0 if damaged | |
1068 */ | |
1069 static int decode_rbsp_trailing(uint8_t *src){ | |
1070 int v= *src; | |
1071 int r; | |
1072 | |
1170 | 1073 tprintf("rbsp trailing %X\n", v); |
1168 | 1074 |
1075 for(r=1; r<9; r++){ | |
1076 if(v&1) return r; | |
1077 v>>=1; | |
1078 } | |
1079 return 0; | |
1080 } | |
1081 | |
1082 /** | |
1083 * idct tranforms the 16 dc values and dequantize them. | |
1084 * @param qp quantization parameter | |
1085 */ | |
1086 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){ | |
1087 const int qmul= dequant_coeff[qp][0]; | |
1088 #define stride 16 | |
1089 int i; | |
1090 int temp[16]; //FIXME check if this is a good idea | |
1091 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride}; | |
1092 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride}; | |
1093 | |
1094 //memset(block, 64, 2*256); | |
1095 //return; | |
1096 for(i=0; i<4; i++){ | |
1097 const int offset= y_offset[i]; | |
1098 const int z0= block[offset+stride*0] + block[offset+stride*4]; | |
1099 const int z1= block[offset+stride*0] - block[offset+stride*4]; | |
1100 const int z2= block[offset+stride*1] - block[offset+stride*5]; | |
1101 const int z3= block[offset+stride*1] + block[offset+stride*5]; | |
1102 | |
1103 temp[4*i+0]= z0+z3; | |
1104 temp[4*i+1]= z1+z2; | |
1105 temp[4*i+2]= z1-z2; | |
1106 temp[4*i+3]= z0-z3; | |
1107 } | |
1108 | |
1109 for(i=0; i<4; i++){ | |
1110 const int offset= x_offset[i]; | |
1111 const int z0= temp[4*0+i] + temp[4*2+i]; | |
1112 const int z1= temp[4*0+i] - temp[4*2+i]; | |
1113 const int z2= temp[4*1+i] - temp[4*3+i]; | |
1114 const int z3= temp[4*1+i] + temp[4*3+i]; | |
1115 | |
1116 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual | |
1117 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2; | |
1118 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2; | |
1119 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2; | |
1120 } | |
1121 } | |
1122 | |
1123 /** | |
1124 * dct tranforms the 16 dc values. | |
1125 * @param qp quantization parameter ??? FIXME | |
1126 */ | |
1127 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){ | |
1128 // const int qmul= dequant_coeff[qp][0]; | |
1129 int i; | |
1130 int temp[16]; //FIXME check if this is a good idea | |
1131 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride}; | |
1132 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride}; | |
1133 | |
1134 for(i=0; i<4; i++){ | |
1135 const int offset= y_offset[i]; | |
1136 const int z0= block[offset+stride*0] + block[offset+stride*4]; | |
1137 const int z1= block[offset+stride*0] - block[offset+stride*4]; | |
1138 const int z2= block[offset+stride*1] - block[offset+stride*5]; | |
1139 const int z3= block[offset+stride*1] + block[offset+stride*5]; | |
1140 | |
1141 temp[4*i+0]= z0+z3; | |
1142 temp[4*i+1]= z1+z2; | |
1143 temp[4*i+2]= z1-z2; | |
1144 temp[4*i+3]= z0-z3; | |
1145 } | |
1146 | |
1147 for(i=0; i<4; i++){ | |
1148 const int offset= x_offset[i]; | |
1149 const int z0= temp[4*0+i] + temp[4*2+i]; | |
1150 const int z1= temp[4*0+i] - temp[4*2+i]; | |
1151 const int z2= temp[4*1+i] - temp[4*3+i]; | |
1152 const int z3= temp[4*1+i] + temp[4*3+i]; | |
1153 | |
1154 block[stride*0 +offset]= (z0 + z3)>>1; | |
1155 block[stride*2 +offset]= (z1 + z2)>>1; | |
1156 block[stride*8 +offset]= (z1 - z2)>>1; | |
1157 block[stride*10+offset]= (z0 - z3)>>1; | |
1158 } | |
1159 } | |
1160 #undef xStride | |
1161 #undef stride | |
1162 | |
1163 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){ | |
1164 const int qmul= dequant_coeff[qp][0]; | |
1165 const int stride= 16*2; | |
1166 const int xStride= 16; | |
1167 int a,b,c,d,e; | |
1168 | |
1169 a= block[stride*0 + xStride*0]; | |
1170 b= block[stride*0 + xStride*1]; | |
1171 c= block[stride*1 + xStride*0]; | |
1172 d= block[stride*1 + xStride*1]; | |
1173 | |
1174 e= a-b; | |
1175 a= a+b; | |
1176 b= c-d; | |
1177 c= c+d; | |
1178 | |
1179 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1; | |
1180 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1; | |
1181 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1; | |
1182 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1; | |
1183 } | |
1184 | |
1185 static void chroma_dc_dct_c(DCTELEM *block){ | |
1186 const int stride= 16*2; | |
1187 const int xStride= 16; | |
1188 int a,b,c,d,e; | |
1189 | |
1190 a= block[stride*0 + xStride*0]; | |
1191 b= block[stride*0 + xStride*1]; | |
1192 c= block[stride*1 + xStride*0]; | |
1193 d= block[stride*1 + xStride*1]; | |
1194 | |
1195 e= a-b; | |
1196 a= a+b; | |
1197 b= c-d; | |
1198 c= c+d; | |
1199 | |
1200 block[stride*0 + xStride*0]= (a+c); | |
1201 block[stride*0 + xStride*1]= (e+b); | |
1202 block[stride*1 + xStride*0]= (a-c); | |
1203 block[stride*1 + xStride*1]= (e-b); | |
1204 } | |
1205 | |
1206 /** | |
1207 * gets the chroma qp. | |
1208 */ | |
1209 static inline int get_chroma_qp(H264Context *h, int qscale){ | |
1210 | |
1211 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)]; | |
1212 } | |
1213 | |
1214 | |
1215 /** | |
1216 * | |
1217 */ | |
1218 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){ | |
1219 int i; | |
1220 uint8_t *cm = cropTbl + MAX_NEG_CROP; | |
1221 | |
1222 block[0] += 32; | |
1223 #if 1 | |
1224 for(i=0; i<4; i++){ | |
1225 const int z0= block[i + 4*0] + block[i + 4*2]; | |
1226 const int z1= block[i + 4*0] - block[i + 4*2]; | |
1227 const int z2= (block[i + 4*1]>>1) - block[i + 4*3]; | |
1228 const int z3= block[i + 4*1] + (block[i + 4*3]>>1); | |
1229 | |
1230 block[i + 4*0]= z0 + z3; | |
1231 block[i + 4*1]= z1 + z2; | |
1232 block[i + 4*2]= z1 - z2; | |
1233 block[i + 4*3]= z0 - z3; | |
1234 } | |
1235 | |
1236 for(i=0; i<4; i++){ | |
1237 const int z0= block[0 + 4*i] + block[2 + 4*i]; | |
1238 const int z1= block[0 + 4*i] - block[2 + 4*i]; | |
1239 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i]; | |
1240 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1); | |
1241 | |
1242 dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ]; | |
1243 dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ]; | |
1244 dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ]; | |
1245 dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ]; | |
1246 } | |
1247 #else | |
1248 for(i=0; i<4; i++){ | |
1249 const int z0= block[0 + 4*i] + block[2 + 4*i]; | |
1250 const int z1= block[0 + 4*i] - block[2 + 4*i]; | |
1251 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i]; | |
1252 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1); | |
1253 | |
1254 block[0 + 4*i]= z0 + z3; | |
1255 block[1 + 4*i]= z1 + z2; | |
1256 block[2 + 4*i]= z1 - z2; | |
1257 block[3 + 4*i]= z0 - z3; | |
1258 } | |
1259 | |
1260 for(i=0; i<4; i++){ | |
1261 const int z0= block[i + 4*0] + block[i + 4*2]; | |
1262 const int z1= block[i + 4*0] - block[i + 4*2]; | |
1263 const int z2= (block[i + 4*1]>>1) - block[i + 4*3]; | |
1264 const int z3= block[i + 4*1] + (block[i + 4*3]>>1); | |
1265 | |
1266 dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ]; | |
1267 dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ]; | |
1268 dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ]; | |
1269 dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ]; | |
1270 } | |
1271 #endif | |
1272 } | |
1273 | |
1274 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){ | |
1275 int i; | |
1276 //FIXME try int temp instead of block | |
1277 | |
1278 for(i=0; i<4; i++){ | |
1279 const int d0= src1[0 + i*stride] - src2[0 + i*stride]; | |
1280 const int d1= src1[1 + i*stride] - src2[1 + i*stride]; | |
1281 const int d2= src1[2 + i*stride] - src2[2 + i*stride]; | |
1282 const int d3= src1[3 + i*stride] - src2[3 + i*stride]; | |
1283 const int z0= d0 + d3; | |
1284 const int z3= d0 - d3; | |
1285 const int z1= d1 + d2; | |
1286 const int z2= d1 - d2; | |
1287 | |
1288 block[0 + 4*i]= z0 + z1; | |
1289 block[1 + 4*i]= 2*z3 + z2; | |
1290 block[2 + 4*i]= z0 - z1; | |
1291 block[3 + 4*i]= z3 - 2*z2; | |
1292 } | |
1293 | |
1294 for(i=0; i<4; i++){ | |
1295 const int z0= block[0*4 + i] + block[3*4 + i]; | |
1296 const int z3= block[0*4 + i] - block[3*4 + i]; | |
1297 const int z1= block[1*4 + i] + block[2*4 + i]; | |
1298 const int z2= block[1*4 + i] - block[2*4 + i]; | |
1299 | |
1300 block[0*4 + i]= z0 + z1; | |
1301 block[1*4 + i]= 2*z3 + z2; | |
1302 block[2*4 + i]= z0 - z1; | |
1303 block[3*4 + i]= z3 - 2*z2; | |
1304 } | |
1305 } | |
1306 | |
1307 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close | |
1308 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away) | |
1309 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){ | |
1310 int i; | |
1311 const int * const quant_table= quant_coeff[qscale]; | |
1312 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6; | |
1313 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1; | |
1314 const unsigned int threshold2= (threshold1<<1); | |
1315 int last_non_zero; | |
1316 | |
1317 if(seperate_dc){ | |
1318 if(qscale<=18){ | |
1319 //avoid overflows | |
1320 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6; | |
1321 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1; | |
1322 const unsigned int dc_threshold2= (dc_threshold1<<1); | |
1323 | |
1324 int level= block[0]*quant_coeff[qscale+18][0]; | |
1325 if(((unsigned)(level+dc_threshold1))>dc_threshold2){ | |
1326 if(level>0){ | |
1327 level= (dc_bias + level)>>(QUANT_SHIFT-2); | |
1328 block[0]= level; | |
1329 }else{ | |
1330 level= (dc_bias - level)>>(QUANT_SHIFT-2); | |
1331 block[0]= -level; | |
1332 } | |
1333 // last_non_zero = i; | |
1334 }else{ | |
1335 block[0]=0; | |
1336 } | |
1337 }else{ | |
1338 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6; | |
1339 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1; | |
1340 const unsigned int dc_threshold2= (dc_threshold1<<1); | |
1341 | |
1342 int level= block[0]*quant_table[0]; | |
1343 if(((unsigned)(level+dc_threshold1))>dc_threshold2){ | |
1344 if(level>0){ | |
1345 level= (dc_bias + level)>>(QUANT_SHIFT+1); | |
1346 block[0]= level; | |
1347 }else{ | |
1348 level= (dc_bias - level)>>(QUANT_SHIFT+1); | |
1349 block[0]= -level; | |
1350 } | |
1351 // last_non_zero = i; | |
1352 }else{ | |
1353 block[0]=0; | |
1354 } | |
1355 } | |
1356 last_non_zero= 0; | |
1357 i=1; | |
1358 }else{ | |
1359 last_non_zero= -1; | |
1360 i=0; | |
1361 } | |
1362 | |
1363 for(; i<16; i++){ | |
1364 const int j= scantable[i]; | |
1365 int level= block[j]*quant_table[j]; | |
1366 | |
1367 // if( bias+level >= (1<<(QMAT_SHIFT - 3)) | |
1368 // || bias-level >= (1<<(QMAT_SHIFT - 3))){ | |
1369 if(((unsigned)(level+threshold1))>threshold2){ | |
1370 if(level>0){ | |
1371 level= (bias + level)>>QUANT_SHIFT; | |
1372 block[j]= level; | |
1373 }else{ | |
1374 level= (bias - level)>>QUANT_SHIFT; | |
1375 block[j]= -level; | |
1376 } | |
1377 last_non_zero = i; | |
1378 }else{ | |
1379 block[j]=0; | |
1380 } | |
1381 } | |
1382 | |
1383 return last_non_zero; | |
1384 } | |
1385 | |
1386 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){ | |
1387 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
1388 ((uint32_t*)(src+0*stride))[0]= a; | |
1389 ((uint32_t*)(src+1*stride))[0]= a; | |
1390 ((uint32_t*)(src+2*stride))[0]= a; | |
1391 ((uint32_t*)(src+3*stride))[0]= a; | |
1392 } | |
1393 | |
1394 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){ | |
1395 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101; | |
1396 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101; | |
1397 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101; | |
1398 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101; | |
1399 } | |
1400 | |
1401 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1402 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] | |
1403 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3; | |
1404 | |
1405 ((uint32_t*)(src+0*stride))[0]= | |
1406 ((uint32_t*)(src+1*stride))[0]= | |
1407 ((uint32_t*)(src+2*stride))[0]= | |
1408 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1409 } | |
1410 | |
1411 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1412 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2; | |
1413 | |
1414 ((uint32_t*)(src+0*stride))[0]= | |
1415 ((uint32_t*)(src+1*stride))[0]= | |
1416 ((uint32_t*)(src+2*stride))[0]= | |
1417 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1418 } | |
1419 | |
1420 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1421 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2; | |
1422 | |
1423 ((uint32_t*)(src+0*stride))[0]= | |
1424 ((uint32_t*)(src+1*stride))[0]= | |
1425 ((uint32_t*)(src+2*stride))[0]= | |
1426 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1427 } | |
1428 | |
1429 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1430 ((uint32_t*)(src+0*stride))[0]= | |
1431 ((uint32_t*)(src+1*stride))[0]= | |
1432 ((uint32_t*)(src+2*stride))[0]= | |
1433 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U; | |
1434 } | |
1435 | |
1436 | |
1437 #define LOAD_TOP_RIGHT_EDGE\ | |
1438 const int t4= topright[0];\ | |
1439 const int t5= topright[1];\ | |
1440 const int t6= topright[2];\ | |
1441 const int t7= topright[3];\ | |
1442 | |
1443 #define LOAD_LEFT_EDGE\ | |
1444 const int l0= src[-1+0*stride];\ | |
1445 const int l1= src[-1+1*stride];\ | |
1446 const int l2= src[-1+2*stride];\ | |
1447 const int l3= src[-1+3*stride];\ | |
1448 | |
1449 #define LOAD_TOP_EDGE\ | |
1450 const int t0= src[ 0-1*stride];\ | |
1451 const int t1= src[ 1-1*stride];\ | |
1452 const int t2= src[ 2-1*stride];\ | |
1453 const int t3= src[ 3-1*stride];\ | |
1454 | |
1455 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){ | |
1456 const int lt= src[-1-1*stride]; | |
1457 LOAD_TOP_EDGE | |
1458 LOAD_LEFT_EDGE | |
1459 | |
1460 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; | |
1461 src[0+2*stride]= | |
1462 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; | |
1463 src[0+1*stride]= | |
1464 src[1+2*stride]= | |
1465 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; | |
1466 src[0+0*stride]= | |
1467 src[1+1*stride]= | |
1468 src[2+2*stride]= | |
1469 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
1470 src[1+0*stride]= | |
1471 src[2+1*stride]= | |
1472 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
1473 src[2+0*stride]= | |
1474 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1475 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1282 | 1476 } |
1168 | 1477 |
1478 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){ | |
1479 LOAD_TOP_EDGE | |
1480 LOAD_TOP_RIGHT_EDGE | |
1481 // LOAD_LEFT_EDGE | |
1482 | |
1483 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2; | |
1484 src[1+0*stride]= | |
1485 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2; | |
1486 src[2+0*stride]= | |
1487 src[1+1*stride]= | |
1488 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2; | |
1489 src[3+0*stride]= | |
1490 src[2+1*stride]= | |
1491 src[1+2*stride]= | |
1492 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2; | |
1493 src[3+1*stride]= | |
1494 src[2+2*stride]= | |
1495 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2; | |
1496 src[3+2*stride]= | |
1497 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2; | |
1498 src[3+3*stride]=(t6 + 3*t7 + 2)>>2; | |
1282 | 1499 } |
1168 | 1500 |
1501 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){ | |
1502 const int lt= src[-1-1*stride]; | |
1503 LOAD_TOP_EDGE | |
1504 LOAD_LEFT_EDGE | |
1505 const __attribute__((unused)) int unu= l3; | |
1506 | |
1507 src[0+0*stride]= | |
1508 src[1+2*stride]=(lt + t0 + 1)>>1; | |
1509 src[1+0*stride]= | |
1510 src[2+2*stride]=(t0 + t1 + 1)>>1; | |
1511 src[2+0*stride]= | |
1512 src[3+2*stride]=(t1 + t2 + 1)>>1; | |
1513 src[3+0*stride]=(t2 + t3 + 1)>>1; | |
1514 src[0+1*stride]= | |
1515 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
1516 src[1+1*stride]= | |
1517 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
1518 src[2+1*stride]= | |
1519 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1520 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1521 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2; | |
1522 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1282 | 1523 } |
1168 | 1524 |
1525 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){ | |
1526 LOAD_TOP_EDGE | |
1527 LOAD_TOP_RIGHT_EDGE | |
1528 const __attribute__((unused)) int unu= t7; | |
1529 | |
1530 src[0+0*stride]=(t0 + t1 + 1)>>1; | |
1531 src[1+0*stride]= | |
1532 src[0+2*stride]=(t1 + t2 + 1)>>1; | |
1533 src[2+0*stride]= | |
1534 src[1+2*stride]=(t2 + t3 + 1)>>1; | |
1535 src[3+0*stride]= | |
1536 src[2+2*stride]=(t3 + t4+ 1)>>1; | |
1537 src[3+2*stride]=(t4 + t5+ 1)>>1; | |
1538 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1539 src[1+1*stride]= | |
1540 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1541 src[2+1*stride]= | |
1542 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2; | |
1543 src[3+1*stride]= | |
1544 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2; | |
1545 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2; | |
1282 | 1546 } |
1168 | 1547 |
1548 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){ | |
1549 LOAD_LEFT_EDGE | |
1550 | |
1551 src[0+0*stride]=(l0 + l1 + 1)>>1; | |
1552 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1553 src[2+0*stride]= | |
1554 src[0+1*stride]=(l1 + l2 + 1)>>1; | |
1555 src[3+0*stride]= | |
1556 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2; | |
1557 src[2+1*stride]= | |
1558 src[0+2*stride]=(l2 + l3 + 1)>>1; | |
1559 src[3+1*stride]= | |
1560 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2; | |
1561 src[3+2*stride]= | |
1562 src[1+3*stride]= | |
1563 src[0+3*stride]= | |
1564 src[2+2*stride]= | |
1565 src[2+3*stride]= | |
1566 src[3+3*stride]=l3; | |
1282 | 1567 } |
1168 | 1568 |
1569 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){ | |
1570 const int lt= src[-1-1*stride]; | |
1571 LOAD_TOP_EDGE | |
1572 LOAD_LEFT_EDGE | |
1573 const __attribute__((unused)) int unu= t3; | |
1574 | |
1575 src[0+0*stride]= | |
1576 src[2+1*stride]=(lt + l0 + 1)>>1; | |
1577 src[1+0*stride]= | |
1578 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
1579 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
1580 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1581 src[0+1*stride]= | |
1582 src[2+2*stride]=(l0 + l1 + 1)>>1; | |
1583 src[1+1*stride]= | |
1584 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2; | |
1585 src[0+2*stride]= | |
1586 src[2+3*stride]=(l1 + l2+ 1)>>1; | |
1587 src[1+2*stride]= | |
1588 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1589 src[0+3*stride]=(l2 + l3 + 1)>>1; | |
1590 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2; | |
1282 | 1591 } |
1168 | 1592 |
1593 static void pred16x16_vertical_c(uint8_t *src, int stride){ | |
1594 int i; | |
1595 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
1596 const uint32_t b= ((uint32_t*)(src-stride))[1]; | |
1597 const uint32_t c= ((uint32_t*)(src-stride))[2]; | |
1598 const uint32_t d= ((uint32_t*)(src-stride))[3]; | |
1599 | |
1600 for(i=0; i<16; i++){ | |
1601 ((uint32_t*)(src+i*stride))[0]= a; | |
1602 ((uint32_t*)(src+i*stride))[1]= b; | |
1603 ((uint32_t*)(src+i*stride))[2]= c; | |
1604 ((uint32_t*)(src+i*stride))[3]= d; | |
1605 } | |
1606 } | |
1607 | |
1608 static void pred16x16_horizontal_c(uint8_t *src, int stride){ | |
1609 int i; | |
1610 | |
1611 for(i=0; i<16; i++){ | |
1612 ((uint32_t*)(src+i*stride))[0]= | |
1613 ((uint32_t*)(src+i*stride))[1]= | |
1614 ((uint32_t*)(src+i*stride))[2]= | |
1615 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101; | |
1616 } | |
1617 } | |
1618 | |
1619 static void pred16x16_dc_c(uint8_t *src, int stride){ | |
1620 int i, dc=0; | |
1621 | |
1622 for(i=0;i<16; i++){ | |
1623 dc+= src[-1+i*stride]; | |
1624 } | |
1625 | |
1626 for(i=0;i<16; i++){ | |
1627 dc+= src[i-stride]; | |
1628 } | |
1629 | |
1630 dc= 0x01010101*((dc + 16)>>5); | |
1631 | |
1632 for(i=0; i<16; i++){ | |
1633 ((uint32_t*)(src+i*stride))[0]= | |
1634 ((uint32_t*)(src+i*stride))[1]= | |
1635 ((uint32_t*)(src+i*stride))[2]= | |
1636 ((uint32_t*)(src+i*stride))[3]= dc; | |
1637 } | |
1638 } | |
1639 | |
1640 static void pred16x16_left_dc_c(uint8_t *src, int stride){ | |
1641 int i, dc=0; | |
1642 | |
1643 for(i=0;i<16; i++){ | |
1644 dc+= src[-1+i*stride]; | |
1645 } | |
1646 | |
1647 dc= 0x01010101*((dc + 8)>>4); | |
1648 | |
1649 for(i=0; i<16; i++){ | |
1650 ((uint32_t*)(src+i*stride))[0]= | |
1651 ((uint32_t*)(src+i*stride))[1]= | |
1652 ((uint32_t*)(src+i*stride))[2]= | |
1653 ((uint32_t*)(src+i*stride))[3]= dc; | |
1654 } | |
1655 } | |
1656 | |
1657 static void pred16x16_top_dc_c(uint8_t *src, int stride){ | |
1658 int i, dc=0; | |
1659 | |
1660 for(i=0;i<16; i++){ | |
1661 dc+= src[i-stride]; | |
1662 } | |
1663 dc= 0x01010101*((dc + 8)>>4); | |
1664 | |
1665 for(i=0; i<16; i++){ | |
1666 ((uint32_t*)(src+i*stride))[0]= | |
1667 ((uint32_t*)(src+i*stride))[1]= | |
1668 ((uint32_t*)(src+i*stride))[2]= | |
1669 ((uint32_t*)(src+i*stride))[3]= dc; | |
1670 } | |
1671 } | |
1672 | |
1673 static void pred16x16_128_dc_c(uint8_t *src, int stride){ | |
1674 int i; | |
1675 | |
1676 for(i=0; i<16; i++){ | |
1677 ((uint32_t*)(src+i*stride))[0]= | |
1678 ((uint32_t*)(src+i*stride))[1]= | |
1679 ((uint32_t*)(src+i*stride))[2]= | |
1680 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U; | |
1681 } | |
1682 } | |
1683 | |
1234 | 1684 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
|
1685 int i, j, k; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1686 int a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1687 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
|
1688 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
|
1689 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
|
1690 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
|
1691 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
|
1692 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
|
1693 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
|
1694 src1 += stride; src2 -= stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1695 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
|
1696 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
|
1697 } |
1234 | 1698 if(svq3){ |
1699 H = ( 5*(H/4) ) / 16; | |
1700 V = ( 5*(V/4) ) / 16; | |
1330
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
1701 |
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
1702 /* required for 100% accuracy */ |
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
1703 i = H; H = V; V = i; |
1234 | 1704 }else{ |
1705 H = ( 5*H+32 ) >> 6; | |
1706 V = ( 5*V+32 ) >> 6; | |
1707 } | |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1708 |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1709 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
|
1710 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
|
1711 int b = a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1712 a += V; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1713 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
|
1714 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
|
1715 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
|
1716 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
|
1717 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
|
1718 b += 4*H; |
1168 | 1719 } |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1720 src += stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1721 } |
1168 | 1722 } |
1723 | |
1234 | 1724 static void pred16x16_plane_c(uint8_t *src, int stride){ |
1725 pred16x16_plane_compat_c(src, stride, 0); | |
1726 } | |
1727 | |
1168 | 1728 static void pred8x8_vertical_c(uint8_t *src, int stride){ |
1729 int i; | |
1730 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
1731 const uint32_t b= ((uint32_t*)(src-stride))[1]; | |
1732 | |
1733 for(i=0; i<8; i++){ | |
1734 ((uint32_t*)(src+i*stride))[0]= a; | |
1735 ((uint32_t*)(src+i*stride))[1]= b; | |
1736 } | |
1737 } | |
1738 | |
1739 static void pred8x8_horizontal_c(uint8_t *src, int stride){ | |
1740 int i; | |
1741 | |
1742 for(i=0; i<8; i++){ | |
1743 ((uint32_t*)(src+i*stride))[0]= | |
1744 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101; | |
1745 } | |
1746 } | |
1747 | |
1748 static void pred8x8_128_dc_c(uint8_t *src, int stride){ | |
1749 int i; | |
1750 | |
1751 for(i=0; i<4; i++){ | |
1752 ((uint32_t*)(src+i*stride))[0]= | |
1753 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U; | |
1754 } | |
1755 for(i=4; i<8; i++){ | |
1756 ((uint32_t*)(src+i*stride))[0]= | |
1757 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U; | |
1758 } | |
1759 } | |
1760 | |
1761 static void pred8x8_left_dc_c(uint8_t *src, int stride){ | |
1762 int i; | |
1763 int dc0, dc2; | |
1764 | |
1765 dc0=dc2=0; | |
1766 for(i=0;i<4; i++){ | |
1767 dc0+= src[-1+i*stride]; | |
1768 dc2+= src[-1+(i+4)*stride]; | |
1769 } | |
1770 dc0= 0x01010101*((dc0 + 2)>>2); | |
1771 dc2= 0x01010101*((dc2 + 2)>>2); | |
1772 | |
1773 for(i=0; i<4; i++){ | |
1774 ((uint32_t*)(src+i*stride))[0]= | |
1775 ((uint32_t*)(src+i*stride))[1]= dc0; | |
1776 } | |
1777 for(i=4; i<8; i++){ | |
1778 ((uint32_t*)(src+i*stride))[0]= | |
1779 ((uint32_t*)(src+i*stride))[1]= dc2; | |
1780 } | |
1781 } | |
1782 | |
1783 static void pred8x8_top_dc_c(uint8_t *src, int stride){ | |
1784 int i; | |
1785 int dc0, dc1; | |
1786 | |
1787 dc0=dc1=0; | |
1788 for(i=0;i<4; i++){ | |
1789 dc0+= src[i-stride]; | |
1790 dc1+= src[4+i-stride]; | |
1791 } | |
1792 dc0= 0x01010101*((dc0 + 2)>>2); | |
1793 dc1= 0x01010101*((dc1 + 2)>>2); | |
1794 | |
1795 for(i=0; i<4; i++){ | |
1796 ((uint32_t*)(src+i*stride))[0]= dc0; | |
1797 ((uint32_t*)(src+i*stride))[1]= dc1; | |
1798 } | |
1799 for(i=4; i<8; i++){ | |
1800 ((uint32_t*)(src+i*stride))[0]= dc0; | |
1801 ((uint32_t*)(src+i*stride))[1]= dc1; | |
1802 } | |
1803 } | |
1804 | |
1805 | |
1806 static void pred8x8_dc_c(uint8_t *src, int stride){ | |
1807 int i; | |
1808 int dc0, dc1, dc2, dc3; | |
1809 | |
1810 dc0=dc1=dc2=0; | |
1811 for(i=0;i<4; i++){ | |
1812 dc0+= src[-1+i*stride] + src[i-stride]; | |
1813 dc1+= src[4+i-stride]; | |
1814 dc2+= src[-1+(i+4)*stride]; | |
1815 } | |
1816 dc3= 0x01010101*((dc1 + dc2 + 4)>>3); | |
1817 dc0= 0x01010101*((dc0 + 4)>>3); | |
1818 dc1= 0x01010101*((dc1 + 2)>>2); | |
1819 dc2= 0x01010101*((dc2 + 2)>>2); | |
1820 | |
1821 for(i=0; i<4; i++){ | |
1822 ((uint32_t*)(src+i*stride))[0]= dc0; | |
1823 ((uint32_t*)(src+i*stride))[1]= dc1; | |
1824 } | |
1825 for(i=4; i<8; i++){ | |
1826 ((uint32_t*)(src+i*stride))[0]= dc2; | |
1827 ((uint32_t*)(src+i*stride))[1]= dc3; | |
1828 } | |
1829 } | |
1830 | |
1831 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
|
1832 int j, k; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1833 int a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1834 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
|
1835 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
|
1836 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
|
1837 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
|
1838 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
|
1839 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
|
1840 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
|
1841 src1 += stride; src2 -= stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1842 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
|
1843 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
|
1844 } |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1845 H = ( 17*H+16 ) >> 5; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1846 V = ( 17*V+16 ) >> 5; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1847 |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1848 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
|
1849 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
|
1850 int b = a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1851 a += V; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1852 src[0] = cm[ (b ) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1853 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
|
1854 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
|
1855 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
|
1856 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
|
1857 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
|
1858 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
|
1859 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
|
1860 src += stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1861 } |
1168 | 1862 } |
1863 | |
1864 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list, | |
1865 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
1866 int src_x_offset, int src_y_offset, | |
1867 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){ | |
1868 MpegEncContext * const s = &h->s; | |
1869 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8; | |
1870 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8; | |
1871 const int luma_xy= (mx&3) + ((my&3)<<2); | |
1872 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize; | |
1873 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize; | |
1874 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize; | |
1875 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it | |
1876 int extra_height= extra_width; | |
1877 int emu=0; | |
1878 const int full_mx= mx>>2; | |
1879 const int full_my= my>>2; | |
1880 | |
1881 assert(pic->data[0]); | |
1882 | |
1883 if(mx&7) extra_width -= 3; | |
1884 if(my&7) extra_height -= 3; | |
1885 | |
1886 if( full_mx < 0-extra_width | |
1887 || full_my < 0-extra_height | |
1888 || full_mx + 16/*FIXME*/ > s->width + extra_width | |
1889 || full_my + 16/*FIXME*/ > s->height + extra_height){ | |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1282
diff
changeset
|
1890 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 | 1891 src_y= s->edge_emu_buffer + 2 + 2*s->linesize; |
1892 emu=1; | |
1893 } | |
1894 | |
1895 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps? | |
1896 if(!square){ | |
1897 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize); | |
1898 } | |
1899 | |
1900 if(s->flags&CODEC_FLAG_GRAY) return; | |
1901 | |
1902 if(emu){ | |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1282
diff
changeset
|
1903 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 | 1904 src_cb= s->edge_emu_buffer; |
1905 } | |
1906 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7); | |
1907 | |
1908 if(emu){ | |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1282
diff
changeset
|
1909 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 | 1910 src_cr= s->edge_emu_buffer; |
1911 } | |
1912 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7); | |
1913 } | |
1914 | |
1915 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta, | |
1916 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
1917 int x_offset, int y_offset, | |
1918 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put, | |
1919 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg, | |
1920 int list0, int list1){ | |
1921 MpegEncContext * const s = &h->s; | |
1922 qpel_mc_func *qpix_op= qpix_put; | |
1923 h264_chroma_mc_func chroma_op= chroma_put; | |
1924 | |
1925 dest_y += 2*x_offset + 2*y_offset*s-> linesize; | |
1926 dest_cb += x_offset + y_offset*s->uvlinesize; | |
1927 dest_cr += x_offset + y_offset*s->uvlinesize; | |
1928 x_offset += 8*s->mb_x; | |
1929 y_offset += 8*s->mb_y; | |
1930 | |
1931 if(list0){ | |
1169 | 1932 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ]; |
1168 | 1933 mc_dir_part(h, ref, n, square, chroma_height, delta, 0, |
1934 dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
1935 qpix_op, chroma_op); | |
1936 | |
1937 qpix_op= qpix_avg; | |
1938 chroma_op= chroma_avg; | |
1939 } | |
1940 | |
1941 if(list1){ | |
1169 | 1942 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ]; |
1168 | 1943 mc_dir_part(h, ref, n, square, chroma_height, delta, 1, |
1944 dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
1945 qpix_op, chroma_op); | |
1946 } | |
1947 } | |
1948 | |
1949 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
1950 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put), | |
1951 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){ | |
1952 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
|
1953 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 1954 const int mb_type= s->current_picture.mb_type[mb_xy]; |
1955 | |
1956 assert(IS_INTER(mb_type)); | |
1957 | |
1958 if(IS_16X16(mb_type)){ | |
1959 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0, | |
1960 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0], | |
1961 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); | |
1962 }else if(IS_16X8(mb_type)){ | |
1963 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0, | |
1964 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0], | |
1965 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); | |
1966 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4, | |
1967 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0], | |
1968 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1)); | |
1969 }else if(IS_8X16(mb_type)){ | |
1970 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0, | |
1971 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], | |
1972 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); | |
1973 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0, | |
1974 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], | |
1975 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1)); | |
1976 }else{ | |
1977 int i; | |
1978 | |
1979 assert(IS_8X8(mb_type)); | |
1980 | |
1981 for(i=0; i<4; i++){ | |
1982 const int sub_mb_type= h->sub_mb_type[i]; | |
1983 const int n= 4*i; | |
1984 int x_offset= (i&1)<<2; | |
1985 int y_offset= (i&2)<<1; | |
1986 | |
1987 if(IS_SUB_8X8(sub_mb_type)){ | |
1988 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
1989 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], | |
1990 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
1991 }else if(IS_SUB_8X4(sub_mb_type)){ | |
1992 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
1993 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1], | |
1994 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
1995 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2, | |
1996 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1], | |
1997 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
1998 }else if(IS_SUB_4X8(sub_mb_type)){ | |
1999 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2000 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], | |
2001 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2002 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset, | |
2003 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], | |
2004 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2005 }else{ | |
2006 int j; | |
2007 assert(IS_SUB_4X4(sub_mb_type)); | |
2008 for(j=0; j<4; j++){ | |
2009 int sub_x_offset= x_offset + 2*(j&1); | |
2010 int sub_y_offset= y_offset + (j&2); | |
2011 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset, | |
2012 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], | |
2013 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2014 } | |
2015 } | |
2016 } | |
2017 } | |
2018 } | |
2019 | |
2020 static void decode_init_vlc(H264Context *h){ | |
2021 static int done = 0; | |
2022 | |
2023 if (!done) { | |
2024 int i; | |
2025 done = 1; | |
2026 | |
2027 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, | |
2028 &chroma_dc_coeff_token_len [0], 1, 1, | |
2029 &chroma_dc_coeff_token_bits[0], 1, 1); | |
2030 | |
2031 for(i=0; i<4; i++){ | |
2032 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, | |
2033 &coeff_token_len [i][0], 1, 1, | |
2034 &coeff_token_bits[i][0], 1, 1); | |
2035 } | |
2036 | |
2037 for(i=0; i<3; i++){ | |
2038 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4, | |
2039 &chroma_dc_total_zeros_len [i][0], 1, 1, | |
2040 &chroma_dc_total_zeros_bits[i][0], 1, 1); | |
2041 } | |
2042 for(i=0; i<15; i++){ | |
2043 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, | |
2044 &total_zeros_len [i][0], 1, 1, | |
2045 &total_zeros_bits[i][0], 1, 1); | |
2046 } | |
2047 | |
2048 for(i=0; i<6; i++){ | |
2049 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, | |
2050 &run_len [i][0], 1, 1, | |
2051 &run_bits[i][0], 1, 1); | |
2052 } | |
2053 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, | |
2054 &run_len [6][0], 1, 1, | |
2055 &run_bits[6][0], 1, 1); | |
2056 } | |
2057 } | |
2058 | |
2059 /** | |
2060 * Sets the intra prediction function pointers. | |
2061 */ | |
2062 static void init_pred_ptrs(H264Context *h){ | |
2063 // MpegEncContext * const s = &h->s; | |
2064 | |
2065 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c; | |
2066 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c; | |
2067 h->pred4x4[DC_PRED ]= pred4x4_dc_c; | |
2068 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c; | |
2069 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c; | |
2070 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c; | |
2071 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c; | |
2072 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c; | |
2073 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c; | |
2074 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c; | |
2075 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c; | |
2076 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c; | |
2077 | |
2078 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c; | |
2079 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c; | |
2080 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c; | |
2081 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c; | |
2082 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c; | |
2083 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c; | |
2084 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c; | |
2085 | |
2086 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c; | |
2087 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c; | |
2088 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c; | |
2089 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c; | |
2090 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c; | |
2091 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c; | |
2092 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c; | |
2093 } | |
2094 | |
2095 static void free_tables(H264Context *h){ | |
2096 av_freep(&h->intra4x4_pred_mode); | |
2097 av_freep(&h->non_zero_count); | |
2098 av_freep(&h->slice_table_base); | |
2099 h->slice_table= NULL; | |
2100 | |
2101 av_freep(&h->mb2b_xy); | |
2102 av_freep(&h->mb2b8_xy); | |
2103 } | |
2104 | |
2105 /** | |
2106 * allocates tables. | |
2107 * needs widzh/height | |
2108 */ | |
2109 static int alloc_tables(H264Context *h){ | |
2110 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
|
2111 const int big_mb_num= s->mb_stride * (s->mb_height+1); |
1168 | 2112 int x,y; |
2113 | |
2114 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t)) | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
2115 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * (16+4+4) * sizeof(uint8_t)) |
1168 | 2116 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t)) |
2117 | |
2118 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
|
2119 h->slice_table= h->slice_table_base + s->mb_stride + 1; |
1168 | 2120 |
2121 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t)); | |
2122 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t)); | |
2123 for(y=0; y<s->mb_height; y++){ | |
2124 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
|
2125 const int mb_xy= x + y*s->mb_stride; |
1168 | 2126 const int b_xy = 4*x + 4*y*h->b_stride; |
2127 const int b8_xy= 2*x + 2*y*h->b8_stride; | |
2128 | |
2129 h->mb2b_xy [mb_xy]= b_xy; | |
2130 h->mb2b8_xy[mb_xy]= b8_xy; | |
2131 } | |
2132 } | |
2133 | |
2134 return 0; | |
2135 fail: | |
2136 free_tables(h); | |
2137 return -1; | |
2138 } | |
2139 | |
2140 static void common_init(H264Context *h){ | |
2141 MpegEncContext * const s = &h->s; | |
2142 | |
2143 s->width = s->avctx->width; | |
2144 s->height = s->avctx->height; | |
2145 s->codec_id= s->avctx->codec->id; | |
2146 | |
2147 init_pred_ptrs(h); | |
2148 | |
1698 | 2149 s->unrestricted_mv=1; |
1168 | 2150 s->decode=1; //FIXME |
2151 } | |
2152 | |
2153 static int decode_init(AVCodecContext *avctx){ | |
2154 H264Context *h= avctx->priv_data; | |
2155 MpegEncContext * const s = &h->s; | |
2156 | |
1892 | 2157 MPV_decode_defaults(s); |
2158 | |
1168 | 2159 s->avctx = avctx; |
2160 common_init(h); | |
2161 | |
2162 s->out_format = FMT_H264; | |
2163 s->workaround_bugs= avctx->workaround_bugs; | |
2164 | |
2165 // set defaults | |
2166 // s->decode_mb= ff_h263_decode_mb; | |
2167 s->low_delay= 1; | |
2168 avctx->pix_fmt= PIX_FMT_YUV420P; | |
2169 | |
2170 decode_init_vlc(h); | |
2171 | |
2172 return 0; | |
2173 } | |
2174 | |
2175 static void frame_start(H264Context *h){ | |
2176 MpegEncContext * const s = &h->s; | |
2177 int i; | |
2178 | |
2179 MPV_frame_start(s, s->avctx); | |
2180 ff_er_frame_start(s); | |
2181 h->mmco_index=0; | |
2182 | |
2183 assert(s->linesize && s->uvlinesize); | |
2184 | |
2185 for(i=0; i<16; i++){ | |
2186 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3); | |
2187 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3); | |
2188 } | |
2189 for(i=0; i<4; i++){ | |
2190 h->block_offset[16+i]= | |
2191 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3); | |
2192 } | |
2193 | |
2194 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1; | |
2195 } | |
2196 | |
2197 static void hl_decode_mb(H264Context *h){ | |
2198 MpegEncContext * const s = &h->s; | |
2199 const int mb_x= s->mb_x; | |
2200 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
|
2201 const int mb_xy= mb_x + mb_y*s->mb_stride; |
1168 | 2202 const int mb_type= s->current_picture.mb_type[mb_xy]; |
2203 uint8_t *dest_y, *dest_cb, *dest_cr; | |
2204 int linesize, uvlinesize /*dct_offset*/; | |
2205 int i; | |
2206 | |
2207 if(!s->decode) | |
2208 return; | |
2209 | |
2210 if(s->mb_skiped){ | |
2211 } | |
2212 | |
2213 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16; | |
2214 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; | |
2215 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; | |
2216 | |
2217 if (h->mb_field_decoding_flag) { | |
2218 linesize = s->linesize * 2; | |
2219 uvlinesize = s->uvlinesize * 2; | |
2220 if(mb_y&1){ //FIXME move out of this func? | |
2221 dest_y -= s->linesize*15; | |
2222 dest_cb-= s->linesize*7; | |
2223 dest_cr-= s->linesize*7; | |
2224 } | |
2225 } else { | |
2226 linesize = s->linesize; | |
2227 uvlinesize = s->uvlinesize; | |
2228 // dct_offset = s->linesize * 16; | |
2229 } | |
2230 | |
2231 if(IS_INTRA(mb_type)){ | |
2232 if(!(s->flags&CODEC_FLAG_GRAY)){ | |
2233 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize); | |
2234 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize); | |
2235 } | |
2236 | |
2237 if(IS_INTRA4x4(mb_type)){ | |
2238 if(!s->encoding){ | |
2239 for(i=0; i<16; i++){ | |
2240 uint8_t * const ptr= dest_y + h->block_offset[i]; | |
2241 uint8_t *topright= ptr + 4 - linesize; | |
2242 const int topright_avail= (h->topright_samples_available<<i)&0x8000; | |
2243 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ]; | |
2244 int tr; | |
2245 | |
2246 if(!topright_avail){ | |
2247 tr= ptr[3 - linesize]*0x01010101; | |
2248 topright= (uint8_t*) &tr; | |
2249 } | |
2250 | |
2251 h->pred4x4[ dir ](ptr, topright, linesize); | |
1234 | 2252 if(h->non_zero_count_cache[ scan8[i] ]){ |
2253 if(s->codec_id == CODEC_ID_H264) | |
2254 h264_add_idct_c(ptr, h->mb + i*16, linesize); | |
2255 else | |
2256 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0); | |
2257 } | |
1168 | 2258 } |
2259 } | |
2260 }else{ | |
2261 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize); | |
1234 | 2262 if(s->codec_id == CODEC_ID_H264) |
2263 h264_luma_dc_dequant_idct_c(h->mb, s->qscale); | |
2264 else | |
2265 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale); | |
1168 | 2266 } |
1234 | 2267 }else if(s->codec_id == CODEC_ID_H264){ |
1168 | 2268 hl_motion(h, dest_y, dest_cb, dest_cr, |
2269 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, | |
2270 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab); | |
2271 } | |
2272 | |
2273 | |
2274 if(!IS_INTRA4x4(mb_type)){ | |
1250 | 2275 if(s->codec_id == CODEC_ID_H264){ |
2276 for(i=0; i<16; i++){ | |
2277 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below | |
2278 uint8_t * const ptr= dest_y + h->block_offset[i]; | |
1234 | 2279 h264_add_idct_c(ptr, h->mb + i*16, linesize); |
1250 | 2280 } |
2281 } | |
2282 }else{ | |
2283 for(i=0; i<16; i++){ | |
2284 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below | |
2285 uint8_t * const ptr= dest_y + h->block_offset[i]; | |
1234 | 2286 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0); |
1250 | 2287 } |
1168 | 2288 } |
2289 } | |
2290 } | |
2291 | |
2292 if(!(s->flags&CODEC_FLAG_GRAY)){ | |
2293 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp); | |
2294 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp); | |
1250 | 2295 if(s->codec_id == CODEC_ID_H264){ |
2296 for(i=16; i<16+4; i++){ | |
2297 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ | |
2298 uint8_t * const ptr= dest_cb + h->block_offset[i]; | |
1234 | 2299 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize); |
1250 | 2300 } |
2301 } | |
2302 for(i=20; i<20+4; i++){ | |
2303 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ | |
2304 uint8_t * const ptr= dest_cr + h->block_offset[i]; | |
2305 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize); | |
2306 } | |
1168 | 2307 } |
1250 | 2308 }else{ |
2309 for(i=16; i<16+4; i++){ | |
2310 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ | |
2311 uint8_t * const ptr= dest_cb + h->block_offset[i]; | |
1234 | 2312 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2); |
1250 | 2313 } |
2314 } | |
2315 for(i=20; i<20+4; i++){ | |
2316 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ | |
2317 uint8_t * const ptr= dest_cr + h->block_offset[i]; | |
2318 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2); | |
2319 } | |
1168 | 2320 } |
2321 } | |
2322 } | |
2323 } | |
2324 | |
2325 static void decode_mb_cabac(H264Context *h){ | |
2326 // MpegEncContext * const s = &h->s; | |
2327 } | |
2328 | |
2329 /** | |
2330 * fills the default_ref_list. | |
2331 */ | |
2332 static int fill_default_ref_list(H264Context *h){ | |
2333 MpegEncContext * const s = &h->s; | |
2334 int i; | |
2335 Picture sorted_short_ref[16]; | |
2336 | |
2337 if(h->slice_type==B_TYPE){ | |
2338 int out_i; | |
2339 int limit= -1; | |
2340 | |
2341 for(out_i=0; out_i<h->short_ref_count; out_i++){ | |
2342 int best_i=-1; | |
2343 int best_poc=-1; | |
2344 | |
2345 for(i=0; i<h->short_ref_count; i++){ | |
2346 const int poc= h->short_ref[i]->poc; | |
2347 if(poc > limit && poc < best_poc){ | |
2348 best_poc= poc; | |
2349 best_i= i; | |
2350 } | |
2351 } | |
2352 | |
2353 assert(best_i != -1); | |
2354 | |
2355 limit= best_poc; | |
2356 sorted_short_ref[out_i]= *h->short_ref[best_i]; | |
2357 } | |
2358 } | |
2359 | |
2360 if(s->picture_structure == PICT_FRAME){ | |
2361 if(h->slice_type==B_TYPE){ | |
2362 const int current_poc= s->current_picture_ptr->poc; | |
2363 int list; | |
2364 | |
2365 for(list=0; list<2; list++){ | |
2366 int index=0; | |
2367 | |
2368 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){ | |
2369 const int i2= list ? h->short_ref_count - i - 1 : i; | |
2370 const int poc= sorted_short_ref[i2].poc; | |
2371 | |
2372 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit | |
2373 | |
2374 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){ | |
2375 h->default_ref_list[list][index ]= sorted_short_ref[i2]; | |
2376 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num; | |
2377 } | |
2378 } | |
2379 | |
2380 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){ | |
2381 if(h->long_ref[i]->reference != 3) continue; | |
2382 | |
2383 h->default_ref_list[ list ][index ]= *h->long_ref[i]; | |
2384 h->default_ref_list[ list ][index++].pic_id= i;; | |
2385 } | |
2386 | |
2387 if(h->long_ref_count > 1 && h->short_ref_count==0){ | |
2388 Picture temp= h->default_ref_list[1][0]; | |
2389 h->default_ref_list[1][0] = h->default_ref_list[1][1]; | |
2390 h->default_ref_list[1][0] = temp; | |
2391 } | |
2392 | |
2393 if(index < h->ref_count[ list ]) | |
2394 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index)); | |
2395 } | |
2396 }else{ | |
2397 int index=0; | |
2398 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){ | |
2399 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit | |
2400 h->default_ref_list[0][index ]= *h->short_ref[i]; | |
2401 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num; | |
2402 } | |
2403 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){ | |
2404 if(h->long_ref[i]->reference != 3) continue; | |
2405 h->default_ref_list[0][index ]= *h->long_ref[i]; | |
2406 h->default_ref_list[0][index++].pic_id= i;; | |
2407 } | |
2408 if(index < h->ref_count[0]) | |
2409 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index)); | |
2410 } | |
2411 }else{ //FIELD | |
2412 if(h->slice_type==B_TYPE){ | |
2413 }else{ | |
2414 //FIXME second field balh | |
2415 } | |
2416 } | |
2417 return 0; | |
2418 } | |
2419 | |
2420 static int decode_ref_pic_list_reordering(H264Context *h){ | |
2421 MpegEncContext * const s = &h->s; | |
2422 int list; | |
2423 | |
2424 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func | |
2425 | |
2426 for(list=0; list<2; list++){ | |
2427 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]); | |
2428 | |
2429 if(get_bits1(&s->gb)){ | |
2430 int pred= h->curr_pic_num; | |
2431 int index; | |
2432 | |
2433 for(index=0; ; index++){ | |
2434 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb); | |
2435 int pic_id; | |
2436 int i; | |
2437 | |
2438 | |
2439 if(index >= h->ref_count[list]){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2440 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n"); |
1168 | 2441 return -1; |
2442 } | |
2443 | |
2444 if(reordering_of_pic_nums_idc<3){ | |
2445 if(reordering_of_pic_nums_idc<2){ | |
2446 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1; | |
2447 | |
2448 if(abs_diff_pic_num >= h->max_pic_num){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2449 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n"); |
1168 | 2450 return -1; |
2451 } | |
2452 | |
2453 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num; | |
2454 else pred+= abs_diff_pic_num; | |
2455 pred &= h->max_pic_num - 1; | |
2456 | |
2457 for(i= h->ref_count[list]-1; i>=index; i--){ | |
2458 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0) | |
2459 break; | |
2460 } | |
2461 }else{ | |
2462 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx | |
2463 | |
2464 for(i= h->ref_count[list]-1; i>=index; i--){ | |
2465 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1) | |
2466 break; | |
2467 } | |
2468 } | |
2469 | |
2470 if(i < index){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2471 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n"); |
1168 | 2472 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME |
2473 }else if(i > index){ | |
2474 Picture tmp= h->ref_list[list][i]; | |
2475 for(; i>index; i--){ | |
2476 h->ref_list[list][i]= h->ref_list[list][i-1]; | |
2477 } | |
2478 h->ref_list[list][index]= tmp; | |
2479 } | |
2480 }else if(reordering_of_pic_nums_idc==3) | |
2481 break; | |
2482 else{ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2483 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n"); |
1168 | 2484 return -1; |
2485 } | |
2486 } | |
2487 } | |
2488 | |
2489 if(h->slice_type!=B_TYPE) break; | |
2490 } | |
2491 return 0; | |
2492 } | |
2493 | |
2494 static int pred_weight_table(H264Context *h){ | |
2495 MpegEncContext * const s = &h->s; | |
2496 int list, i; | |
2497 | |
2498 h->luma_log2_weight_denom= get_ue_golomb(&s->gb); | |
2499 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb); | |
2500 | |
2501 for(list=0; list<2; list++){ | |
2502 for(i=0; i<h->ref_count[list]; i++){ | |
2503 int luma_weight_flag, chroma_weight_flag; | |
2504 | |
2505 luma_weight_flag= get_bits1(&s->gb); | |
2506 if(luma_weight_flag){ | |
2507 h->luma_weight[list][i]= get_se_golomb(&s->gb); | |
2508 h->luma_offset[list][i]= get_se_golomb(&s->gb); | |
2509 } | |
2510 | |
2511 chroma_weight_flag= get_bits1(&s->gb); | |
2512 if(chroma_weight_flag){ | |
2513 int j; | |
2514 for(j=0; j<2; j++){ | |
2515 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb); | |
2516 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb); | |
2517 } | |
2518 } | |
2519 } | |
2520 if(h->slice_type != B_TYPE) break; | |
2521 } | |
2522 return 0; | |
2523 } | |
2524 | |
2525 /** | |
2526 * instantaneos decoder refresh. | |
2527 */ | |
2528 static void idr(H264Context *h){ | |
2529 int i; | |
2530 | |
2531 for(i=0; i<h->long_ref_count; i++){ | |
2532 h->long_ref[i]->reference=0; | |
2533 h->long_ref[i]= NULL; | |
2534 } | |
2535 h->long_ref_count=0; | |
2536 | |
2537 for(i=0; i<h->short_ref_count; i++){ | |
2538 h->short_ref[i]->reference=0; | |
2539 h->short_ref[i]= NULL; | |
2540 } | |
2541 h->short_ref_count=0; | |
2542 } | |
2543 | |
2544 /** | |
2545 * | |
2546 * @return the removed picture or NULL if an error occures | |
2547 */ | |
2548 static Picture * remove_short(H264Context *h, int frame_num){ | |
1169 | 2549 MpegEncContext * const s = &h->s; |
1168 | 2550 int i; |
2551 | |
1169 | 2552 if(s->avctx->debug&FF_DEBUG_MMCO) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2553 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count); |
1169 | 2554 |
1168 | 2555 for(i=0; i<h->short_ref_count; i++){ |
2556 Picture *pic= h->short_ref[i]; | |
1169 | 2557 if(s->avctx->debug&FF_DEBUG_MMCO) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2558 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic); |
1168 | 2559 if(pic->frame_num == frame_num){ |
2560 h->short_ref[i]= NULL; | |
2561 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*)); | |
2562 h->short_ref_count--; | |
2563 return pic; | |
2564 } | |
2565 } | |
2566 return NULL; | |
2567 } | |
2568 | |
2569 /** | |
2570 * | |
2571 * @return the removed picture or NULL if an error occures | |
2572 */ | |
2573 static Picture * remove_long(H264Context *h, int i){ | |
2574 Picture *pic; | |
2575 | |
2576 if(i >= h->long_ref_count) return NULL; | |
2577 pic= h->long_ref[i]; | |
2578 if(pic==NULL) return NULL; | |
2579 | |
2580 h->long_ref[i]= NULL; | |
2581 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*)); | |
2582 h->long_ref_count--; | |
2583 | |
2584 return pic; | |
2585 } | |
2586 | |
2587 /** | |
2588 * Executes the reference picture marking (memory management control operations). | |
2589 */ | |
2590 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){ | |
2591 MpegEncContext * const s = &h->s; | |
2592 int i; | |
2593 int current_is_long=0; | |
2594 Picture *pic; | |
2595 | |
2596 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0) | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2597 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n"); |
1168 | 2598 |
2599 for(i=0; i<mmco_count; i++){ | |
2600 if(s->avctx->debug&FF_DEBUG_MMCO) | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2601 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index); |
1168 | 2602 |
2603 switch(mmco[i].opcode){ | |
2604 case MMCO_SHORT2UNUSED: | |
2605 pic= remove_short(h, mmco[i].short_frame_num); | |
2606 if(pic==NULL) return -1; | |
2607 pic->reference= 0; | |
2608 break; | |
2609 case MMCO_SHORT2LONG: | |
2610 pic= remove_long(h, mmco[i].long_index); | |
2611 if(pic) pic->reference=0; | |
2612 | |
2613 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num); | |
2614 h->long_ref[ mmco[i].long_index ]->long_ref=1; | |
2615 break; | |
2616 case MMCO_LONG2UNUSED: | |
2617 pic= remove_long(h, mmco[i].long_index); | |
2618 if(pic==NULL) return -1; | |
2619 pic->reference= 0; | |
2620 break; | |
2621 case MMCO_LONG: | |
2622 pic= remove_long(h, mmco[i].long_index); | |
2623 if(pic) pic->reference=0; | |
2624 | |
2625 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr; | |
2626 h->long_ref[ mmco[i].long_index ]->long_ref=1; | |
2627 h->long_ref_count++; | |
2628 | |
2629 current_is_long=1; | |
2630 break; | |
2631 case MMCO_SET_MAX_LONG: | |
2632 assert(mmco[i].long_index <= 16); | |
2633 while(mmco[i].long_index < h->long_ref_count){ | |
2634 pic= remove_long(h, mmco[i].long_index); | |
2635 pic->reference=0; | |
2636 } | |
2637 while(mmco[i].long_index > h->long_ref_count){ | |
2638 h->long_ref[ h->long_ref_count++ ]= NULL; | |
2639 } | |
2640 break; | |
2641 case MMCO_RESET: | |
2642 while(h->short_ref_count){ | |
2643 pic= remove_short(h, h->short_ref[0]->frame_num); | |
2644 pic->reference=0; | |
2645 } | |
2646 while(h->long_ref_count){ | |
2647 pic= remove_long(h, h->long_ref_count-1); | |
2648 pic->reference=0; | |
2649 } | |
2650 break; | |
2651 default: assert(0); | |
2652 } | |
2653 } | |
2654 | |
2655 if(!current_is_long){ | |
2656 pic= remove_short(h, s->current_picture_ptr->frame_num); | |
2657 if(pic){ | |
2658 pic->reference=0; | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2659 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n"); |
1168 | 2660 } |
2661 | |
2662 if(h->short_ref_count) | |
1169 | 2663 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*)); |
2664 | |
2665 h->short_ref[0]= s->current_picture_ptr; | |
1168 | 2666 h->short_ref[0]->long_ref=0; |
2667 h->short_ref_count++; | |
2668 } | |
2669 | |
2670 return 0; | |
2671 } | |
2672 | |
2673 static int decode_ref_pic_marking(H264Context *h){ | |
2674 MpegEncContext * const s = &h->s; | |
2675 int i; | |
2676 | |
2677 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields | |
2678 s->broken_link= get_bits1(&s->gb) -1; | |
2679 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx | |
2680 if(h->mmco[0].long_index == -1) | |
2681 h->mmco_index= 0; | |
2682 else{ | |
2683 h->mmco[0].opcode= MMCO_LONG; | |
2684 h->mmco_index= 1; | |
2685 } | |
2686 }else{ | |
2687 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag | |
2688 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) { | |
2689 MMCOOpcode opcode= get_ue_golomb(&s->gb);; | |
2690 | |
2691 h->mmco[i].opcode= opcode; | |
2692 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){ | |
2693 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 | |
2694 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){ | |
2695 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco); | |
2696 return -1; | |
2697 }*/ | |
2698 } | |
2699 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){ | |
2700 h->mmco[i].long_index= get_ue_golomb(&s->gb); | |
2701 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){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2702 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode); |
1168 | 2703 return -1; |
2704 } | |
2705 } | |
2706 | |
2707 if(opcode > MMCO_LONG){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2708 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode); |
1168 | 2709 return -1; |
2710 } | |
2711 } | |
2712 h->mmco_index= i; | |
2713 }else{ | |
2714 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count); | |
2715 | |
2716 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields | |
2717 h->mmco[0].opcode= MMCO_SHORT2UNUSED; | |
2718 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num; | |
2719 h->mmco_index= 1; | |
2720 }else | |
2721 h->mmco_index= 0; | |
2722 } | |
2723 } | |
2724 | |
2725 return 0; | |
2726 } | |
2727 | |
2728 static int init_poc(H264Context *h){ | |
2729 MpegEncContext * const s = &h->s; | |
2730 const int max_frame_num= 1<<h->sps.log2_max_frame_num; | |
2731 int field_poc[2]; | |
2732 | |
2733 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
2734 h->frame_num_offset= 0; | |
2735 }else{ | |
2736 if(h->frame_num < h->prev_frame_num) | |
2737 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num; | |
2738 else | |
2739 h->frame_num_offset= h->prev_frame_num_offset; | |
2740 } | |
2741 | |
2742 if(h->sps.poc_type==0){ | |
2743 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb; | |
2744 | |
2745 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2) | |
2746 h->poc_msb = h->prev_poc_msb + max_poc_lsb; | |
2747 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2) | |
2748 h->poc_msb = h->prev_poc_msb - max_poc_lsb; | |
2749 else | |
2750 h->poc_msb = h->prev_poc_msb; | |
2751 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb); | |
2752 field_poc[0] = | |
2753 field_poc[1] = h->poc_msb + h->poc_lsb; | |
2754 if(s->picture_structure == PICT_FRAME) | |
2755 field_poc[1] += h->delta_poc_bottom; | |
2756 }else if(h->sps.poc_type==1){ | |
2757 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc; | |
2758 int i; | |
2759 | |
2760 if(h->sps.poc_cycle_length != 0) | |
2761 abs_frame_num = h->frame_num_offset + h->frame_num; | |
2762 else | |
2763 abs_frame_num = 0; | |
2764 | |
2765 if(h->nal_ref_idc==0 && abs_frame_num > 0) | |
2766 abs_frame_num--; | |
2767 | |
2768 expected_delta_per_poc_cycle = 0; | |
2769 for(i=0; i < h->sps.poc_cycle_length; i++) | |
2770 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse | |
2771 | |
2772 if(abs_frame_num > 0){ | |
2773 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length; | |
2774 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length; | |
2775 | |
2776 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle; | |
2777 for(i = 0; i <= frame_num_in_poc_cycle; i++) | |
2778 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ]; | |
2779 } else | |
2780 expectedpoc = 0; | |
2781 | |
2782 if(h->nal_ref_idc == 0) | |
2783 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic; | |
2784 | |
2785 field_poc[0] = expectedpoc + h->delta_poc[0]; | |
2786 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field; | |
2787 | |
2788 if(s->picture_structure == PICT_FRAME) | |
2789 field_poc[1] += h->delta_poc[1]; | |
2790 }else{ | |
2791 int poc; | |
2792 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
2793 poc= 0; | |
2794 }else{ | |
2795 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num); | |
2796 else poc= 2*(h->frame_num_offset + h->frame_num) - 1; | |
2797 } | |
2798 field_poc[0]= poc; | |
2799 field_poc[1]= poc; | |
2800 } | |
2801 | |
2802 if(s->picture_structure != PICT_BOTTOM_FIELD) | |
2803 s->current_picture_ptr->field_poc[0]= field_poc[0]; | |
2804 if(s->picture_structure != PICT_TOP_FIELD) | |
2805 s->current_picture_ptr->field_poc[1]= field_poc[1]; | |
2806 if(s->picture_structure == PICT_FRAME) // FIXME field pix? | |
2807 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]); | |
2808 | |
2809 return 0; | |
2810 } | |
2811 | |
2812 /** | |
2813 * decodes a slice header. | |
2814 * this will allso call MPV_common_init() and frame_start() as needed | |
2815 */ | |
2816 static int decode_slice_header(H264Context *h){ | |
2817 MpegEncContext * const s = &h->s; | |
2818 int first_mb_in_slice, pps_id; | |
2819 int num_ref_idx_active_override_flag; | |
2820 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE}; | |
2821 | |
2822 s->current_picture.reference= h->nal_ref_idc != 0; | |
2823 | |
2824 first_mb_in_slice= get_ue_golomb(&s->gb); | |
2825 | |
2826 h->slice_type= get_ue_golomb(&s->gb); | |
2827 if(h->slice_type > 9){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2828 av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y); |
1168 | 2829 } |
2830 if(h->slice_type > 4){ | |
2831 h->slice_type -= 5; | |
2832 h->slice_type_fixed=1; | |
2833 }else | |
2834 h->slice_type_fixed=0; | |
2835 | |
2836 h->slice_type= slice_type_map[ h->slice_type ]; | |
2837 | |
2838 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though | |
2839 | |
2840 pps_id= get_ue_golomb(&s->gb); | |
2841 if(pps_id>255){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2842 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n"); |
1168 | 2843 return -1; |
2844 } | |
2845 h->pps= h->pps_buffer[pps_id]; | |
1174 | 2846 if(h->pps.slice_group_count == 0){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2847 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n"); |
1174 | 2848 return -1; |
2849 } | |
2850 | |
1168 | 2851 h->sps= h->sps_buffer[ h->pps.sps_id ]; |
1174 | 2852 if(h->sps.log2_max_frame_num == 0){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2853 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n"); |
1174 | 2854 return -1; |
2855 } | |
1168 | 2856 |
2857 s->mb_width= h->sps.mb_width; | |
2858 s->mb_height= h->sps.mb_height; | |
2859 | |
2860 h->b_stride= s->mb_width*4; | |
2861 h->b8_stride= s->mb_width*2; | |
2862 | |
2863 s->mb_x = first_mb_in_slice % s->mb_width; | |
2864 s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW | |
2865 | |
1371 | 2866 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right ); |
1168 | 2867 if(h->sps.frame_mbs_only_flag) |
1371 | 2868 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom); |
1168 | 2869 else |
1371 | 2870 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck |
1168 | 2871 |
2872 if (s->context_initialized | |
1548 | 2873 && ( s->width != s->avctx->width || s->height != s->avctx->height)) { |
1168 | 2874 free_tables(h); |
2875 MPV_common_end(s); | |
2876 } | |
2877 if (!s->context_initialized) { | |
2878 if (MPV_common_init(s) < 0) | |
2879 return -1; | |
2880 | |
2881 alloc_tables(h); | |
2882 | |
2883 s->avctx->width = s->width; | |
2884 s->avctx->height = s->height; | |
1548 | 2885 s->avctx->sample_aspect_ratio= h->sps.sar; |
1168 | 2886 } |
2887 | |
2888 if(first_mb_in_slice == 0){ | |
2889 frame_start(h); | |
2890 } | |
2891 | |
1169 | 2892 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup |
1168 | 2893 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num); |
2894 | |
2895 if(h->sps.frame_mbs_only_flag){ | |
2896 s->picture_structure= PICT_FRAME; | |
2897 }else{ | |
2898 if(get_bits1(&s->gb)) //field_pic_flag | |
2899 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag | |
2900 else | |
2901 s->picture_structure= PICT_FRAME; | |
2902 } | |
2903 | |
2904 if(s->picture_structure==PICT_FRAME){ | |
2905 h->curr_pic_num= h->frame_num; | |
2906 h->max_pic_num= 1<< h->sps.log2_max_frame_num; | |
2907 }else{ | |
2908 h->curr_pic_num= 2*h->frame_num; | |
2909 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1); | |
2910 } | |
2911 | |
2912 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
1453 | 2913 get_ue_golomb(&s->gb); /* idr_pic_id */ |
1168 | 2914 } |
2915 | |
2916 if(h->sps.poc_type==0){ | |
2917 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb); | |
2918 | |
2919 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){ | |
2920 h->delta_poc_bottom= get_se_golomb(&s->gb); | |
2921 } | |
2922 } | |
2923 | |
2924 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){ | |
2925 h->delta_poc[0]= get_se_golomb(&s->gb); | |
2926 | |
2927 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME) | |
2928 h->delta_poc[1]= get_se_golomb(&s->gb); | |
2929 } | |
2930 | |
2931 init_poc(h); | |
2932 | |
2933 if(h->pps.redundant_pic_cnt_present){ | |
2934 h->redundant_pic_count= get_ue_golomb(&s->gb); | |
2935 } | |
2936 | |
2937 //set defaults, might be overriden a few line later | |
2938 h->ref_count[0]= h->pps.ref_count[0]; | |
2939 h->ref_count[1]= h->pps.ref_count[1]; | |
2940 | |
2941 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){ | |
2942 if(h->slice_type == B_TYPE){ | |
2943 h->direct_spatial_mv_pred= get_bits1(&s->gb); | |
2944 } | |
2945 num_ref_idx_active_override_flag= get_bits1(&s->gb); | |
2946 | |
2947 if(num_ref_idx_active_override_flag){ | |
2948 h->ref_count[0]= get_ue_golomb(&s->gb) + 1; | |
2949 if(h->slice_type==B_TYPE) | |
2950 h->ref_count[1]= get_ue_golomb(&s->gb) + 1; | |
2951 | |
2952 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2953 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n"); |
1168 | 2954 return -1; |
2955 } | |
2956 } | |
2957 } | |
2958 | |
2959 if(first_mb_in_slice == 0){ | |
2960 fill_default_ref_list(h); | |
2961 } | |
2962 | |
2963 decode_ref_pic_list_reordering(h); | |
2964 | |
2965 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) | |
2966 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) ) | |
2967 pred_weight_table(h); | |
2968 | |
2969 if(s->current_picture.reference) | |
2970 decode_ref_pic_marking(h); | |
2971 //FIXME CABAC stuff | |
2972 | |
2973 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta | |
1898 | 2974 if(s->qscale<0 || s->qscale>51){ |
2975 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale); | |
2976 return -1; | |
2977 } | |
1168 | 2978 //FIXME qscale / qp ... stuff |
2979 if(h->slice_type == SP_TYPE){ | |
1453 | 2980 get_bits1(&s->gb); /* sp_for_switch_flag */ |
1168 | 2981 } |
2982 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){ | |
1453 | 2983 get_se_golomb(&s->gb); /* slice_qs_delta */ |
1168 | 2984 } |
2985 | |
1898 | 2986 h->disable_deblocking_filter_idc = 0; |
2987 h->slice_alpha_c0_offset = 0; | |
2988 h->slice_beta_offset = 0; | |
1168 | 2989 if( h->pps.deblocking_filter_parameters_present ) { |
2990 h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb); | |
2991 if( h->disable_deblocking_filter_idc != 1 ) { | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
2992 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
2993 h->slice_beta_offset = get_se_golomb(&s->gb) << 1; |
1168 | 2994 } |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
2995 } |
1168 | 2996 |
2997 #if 0 //FMO | |
2998 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5) | |
2999 slice_group_change_cycle= get_bits(&s->gb, ?); | |
3000 #endif | |
3001 | |
3002 if(s->avctx->debug&FF_DEBUG_PICT_INFO){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3003 av_log(h->s.avctx, AV_LOG_DEBUG, "mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n", |
1168 | 3004 first_mb_in_slice, |
1264 | 3005 av_get_pict_type_char(h->slice_type), |
1168 | 3006 pps_id, h->frame_num, |
3007 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1], | |
3008 h->ref_count[0], h->ref_count[1], | |
3009 s->qscale, | |
3010 h->disable_deblocking_filter_idc | |
3011 ); | |
3012 } | |
3013 | |
3014 return 0; | |
3015 } | |
3016 | |
3017 /** | |
3018 * | |
3019 */ | |
3020 static inline int get_level_prefix(GetBitContext *gb){ | |
3021 unsigned int buf; | |
3022 int log; | |
3023 | |
3024 OPEN_READER(re, gb); | |
3025 UPDATE_CACHE(re, gb); | |
3026 buf=GET_CACHE(re, gb); | |
3027 | |
3028 log= 32 - av_log2(buf); | |
3029 #ifdef TRACE | |
3030 print_bin(buf>>(32-log), log); | |
3031 printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__); | |
3032 #endif | |
3033 | |
3034 LAST_SKIP_BITS(re, gb, log); | |
3035 CLOSE_READER(re, gb); | |
3036 | |
3037 return log-1; | |
3038 } | |
3039 | |
3040 /** | |
3041 * decodes a residual block. | |
3042 * @param n block index | |
3043 * @param scantable scantable | |
3044 * @param max_coeff number of coefficients in the block | |
3045 * @return <0 if an error occured | |
3046 */ | |
3047 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){ | |
3048 MpegEncContext * const s = &h->s; | |
3049 const uint16_t *qmul= dequant_coeff[qp]; | |
3050 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}; | |
3051 int level[16], run[16]; | |
3052 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones; | |
3053 | |
3054 //FIXME put trailing_onex into the context | |
3055 | |
3056 if(n == CHROMA_DC_BLOCK_INDEX){ | |
3057 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1); | |
3058 total_coeff= coeff_token>>2; | |
3059 }else{ | |
3060 if(n == LUMA_DC_BLOCK_INDEX){ | |
3061 total_coeff= pred_non_zero_count(h, 0); | |
3062 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2); | |
3063 total_coeff= coeff_token>>2; | |
3064 }else{ | |
3065 total_coeff= pred_non_zero_count(h, n); | |
3066 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2); | |
3067 total_coeff= coeff_token>>2; | |
3068 h->non_zero_count_cache[ scan8[n] ]= total_coeff; | |
3069 } | |
3070 } | |
3071 | |
3072 //FIXME set last_non_zero? | |
3073 | |
3074 if(total_coeff==0) | |
3075 return 0; | |
3076 | |
3077 trailing_ones= coeff_token&3; | |
1170 | 3078 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff); |
1168 | 3079 assert(total_coeff<=16); |
3080 | |
3081 for(i=0; i<trailing_ones; i++){ | |
3082 level[i]= 1 - 2*get_bits1(gb); | |
3083 } | |
3084 | |
3085 suffix_length= total_coeff > 10 && trailing_ones < 3; | |
3086 | |
3087 for(; i<total_coeff; i++){ | |
3088 const int prefix= get_level_prefix(gb); | |
3089 int level_code, mask; | |
3090 | |
3091 if(prefix<14){ //FIXME try to build a large unified VLC table for all this | |
3092 if(suffix_length) | |
3093 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part | |
3094 else | |
3095 level_code= (prefix<<suffix_length); //part | |
3096 }else if(prefix==14){ | |
3097 if(suffix_length) | |
3098 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part | |
3099 else | |
3100 level_code= prefix + get_bits(gb, 4); //part | |
3101 }else if(prefix==15){ | |
3102 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part | |
3103 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense | |
3104 }else{ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3105 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y); |
1168 | 3106 return -1; |
3107 } | |
3108 | |
3109 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration | |
3110 | |
3111 mask= -(level_code&1); | |
3112 level[i]= (((2+level_code)>>1) ^ mask) - mask; | |
3113 | |
3114 if(suffix_length==0) suffix_length=1; //FIXME split first iteration | |
3115 | |
3116 #if 1 | |
3117 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++; | |
3118 #else | |
3119 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++; | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3120 /* ? == prefix > 2 or sth */ |
1168 | 3121 #endif |
1170 | 3122 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length); |
1168 | 3123 } |
3124 | |
3125 if(total_coeff == max_coeff) | |
3126 zeros_left=0; | |
3127 else{ | |
3128 if(n == CHROMA_DC_BLOCK_INDEX) | |
3129 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1); | |
3130 else | |
3131 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1); | |
3132 } | |
3133 | |
3134 for(i=0; i<total_coeff-1; i++){ | |
3135 if(zeros_left <=0) | |
3136 break; | |
3137 else if(zeros_left < 7){ | |
3138 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1); | |
3139 }else{ | |
3140 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); | |
3141 } | |
3142 zeros_left -= run[i]; | |
3143 } | |
3144 | |
3145 if(zeros_left<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3146 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y); |
1168 | 3147 return -1; |
3148 } | |
3149 | |
3150 for(; i<total_coeff-1; i++){ | |
3151 run[i]= 0; | |
3152 } | |
3153 | |
3154 run[i]= zeros_left; | |
3155 | |
3156 coeff_num=-1; | |
3157 if(n > 24){ | |
3158 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode? | |
3159 int j; | |
3160 | |
3161 coeff_num += run[i] + 1; //FIXME add 1 earlier ? | |
3162 j= scantable[ coeff_num ]; | |
3163 | |
3164 block[j]= level[i]; | |
3165 } | |
3166 }else{ | |
3167 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode? | |
3168 int j; | |
3169 | |
3170 coeff_num += run[i] + 1; //FIXME add 1 earlier ? | |
3171 j= scantable[ coeff_num ]; | |
3172 | |
3173 block[j]= level[i] * qmul[j]; | |
3174 // printf("%d %d ", block[j], qmul[j]); | |
3175 } | |
3176 } | |
3177 return 0; | |
3178 } | |
3179 | |
3180 /** | |
3181 * decodes a macroblock | |
3182 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed | |
3183 */ | |
3184 static int decode_mb(H264Context *h){ | |
3185 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
|
3186 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1169 | 3187 int mb_type, partition_count, cbp; |
1168 | 3188 |
1252 | 3189 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong? |
1168 | 3190 |
1170 | 3191 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y); |
1453 | 3192 cbp = 0; /* avoid warning. FIXME: find a solution without slowing |
3193 down the code */ | |
1168 | 3194 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){ |
3195 if(s->mb_skip_run==-1) | |
3196 s->mb_skip_run= get_ue_golomb(&s->gb); | |
3197 | |
3198 if (s->mb_skip_run--) { | |
1187 | 3199 int mx, my; |
1168 | 3200 /* skip mb */ |
3201 //FIXME b frame | |
3202 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0; | |
3203 | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3204 memset(h->non_zero_count[mb_xy], 0, 16+4+4); |
1168 | 3205 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui |
3206 | |
3207 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){ | |
3208 h->mb_field_decoding_flag= get_bits1(&s->gb); | |
3209 } | |
3210 | |
3211 if(h->mb_field_decoding_flag) | |
3212 mb_type|= MB_TYPE_INTERLACED; | |
3213 | |
3214 fill_caches(h, mb_type); //FIXME check what is needed and what not ... | |
3215 pred_pskip_motion(h, &mx, &my); | |
3216 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1); | |
1269 | 3217 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4); |
1168 | 3218 write_back_motion(h, mb_type); |
3219 | |
3220 s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3221 s->current_picture.qscale_table[mb_xy]= s->qscale; |
1168 | 3222 h->slice_table[ mb_xy ]= h->slice_num; |
3223 | |
3224 h->prev_mb_skiped= 1; | |
3225 return 0; | |
3226 } | |
3227 } | |
3228 if(h->sps.mb_aff /* && !field pic FIXME needed? */){ | |
3229 if((s->mb_y&1)==0) | |
3230 h->mb_field_decoding_flag = get_bits1(&s->gb); | |
3231 }else | |
3232 h->mb_field_decoding_flag=0; //FIXME som ed note ?! | |
3233 | |
3234 h->prev_mb_skiped= 0; | |
3235 | |
3236 mb_type= get_ue_golomb(&s->gb); | |
3237 if(h->slice_type == B_TYPE){ | |
3238 if(mb_type < 23){ | |
3239 partition_count= b_mb_type_info[mb_type].partition_count; | |
3240 mb_type= b_mb_type_info[mb_type].type; | |
3241 }else{ | |
3242 mb_type -= 23; | |
3243 goto decode_intra_mb; | |
3244 } | |
3245 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){ | |
3246 if(mb_type < 5){ | |
3247 partition_count= p_mb_type_info[mb_type].partition_count; | |
3248 mb_type= p_mb_type_info[mb_type].type; | |
3249 }else{ | |
3250 mb_type -= 5; | |
3251 goto decode_intra_mb; | |
3252 } | |
3253 }else{ | |
3254 assert(h->slice_type == I_TYPE); | |
3255 decode_intra_mb: | |
3256 if(mb_type > 25){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3257 av_log(h->s.avctx, AV_LOG_ERROR, "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 | 3258 return -1; |
3259 } | |
3260 partition_count=0; | |
3261 cbp= i_mb_type_info[mb_type].cbp; | |
3262 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode; | |
3263 mb_type= i_mb_type_info[mb_type].type; | |
3264 } | |
3265 | |
3266 if(h->mb_field_decoding_flag) | |
3267 mb_type |= MB_TYPE_INTERLACED; | |
3268 | |
3269 s->current_picture.mb_type[mb_xy]= mb_type; | |
3270 h->slice_table[ mb_xy ]= h->slice_num; | |
3271 | |
3272 if(IS_INTRA_PCM(mb_type)){ | |
3273 const uint8_t *ptr; | |
1187 | 3274 int x, y; |
1168 | 3275 |
3276 // we assume these blocks are very rare so we dont optimize it | |
3277 align_get_bits(&s->gb); | |
3278 | |
3279 ptr= s->gb.buffer + get_bits_count(&s->gb); | |
3280 | |
3281 for(y=0; y<16; y++){ | |
3282 const int index= 4*(y&3) + 64*(y>>2); | |
3283 for(x=0; x<16; x++){ | |
3284 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++); | |
3285 } | |
3286 } | |
3287 for(y=0; y<8; y++){ | |
3288 const int index= 256 + 4*(y&3) + 32*(y>>2); | |
3289 for(x=0; x<8; x++){ | |
3290 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++); | |
3291 } | |
3292 } | |
3293 for(y=0; y<8; y++){ | |
3294 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2); | |
3295 for(x=0; x<8; x++){ | |
3296 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++); | |
3297 } | |
3298 } | |
3299 | |
3300 skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers | |
3301 | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3302 memset(h->non_zero_count[mb_xy], 16, 16+4+4); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3303 s->current_picture.qscale_table[mb_xy]= s->qscale; |
1168 | 3304 |
3305 return 0; | |
3306 } | |
3307 | |
3308 fill_caches(h, mb_type); | |
3309 | |
3310 //mb_pred | |
3311 if(IS_INTRA(mb_type)){ | |
3312 // init_top_left_availability(h); | |
3313 if(IS_INTRA4x4(mb_type)){ | |
3314 int i; | |
3315 | |
3316 // fill_intra4x4_pred_table(h); | |
3317 for(i=0; i<16; i++){ | |
3318 const int mode_coded= !get_bits1(&s->gb); | |
3319 const int predicted_mode= pred_intra_mode(h, i); | |
3320 int mode; | |
3321 | |
3322 if(mode_coded){ | |
3323 const int rem_mode= get_bits(&s->gb, 3); | |
3324 if(rem_mode<predicted_mode) | |
3325 mode= rem_mode; | |
3326 else | |
3327 mode= rem_mode + 1; | |
3328 }else{ | |
3329 mode= predicted_mode; | |
3330 } | |
3331 | |
3332 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode; | |
3333 } | |
3334 write_back_intra_pred_mode(h); | |
3335 if( check_intra4x4_pred_mode(h) < 0) | |
3336 return -1; | |
3337 }else{ | |
3338 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode); | |
3339 if(h->intra16x16_pred_mode < 0) | |
3340 return -1; | |
3341 } | |
3342 h->chroma_pred_mode= get_ue_golomb(&s->gb); | |
3343 | |
3344 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode); | |
3345 if(h->chroma_pred_mode < 0) | |
3346 return -1; | |
3347 }else if(partition_count==4){ | |
3348 int i, j, sub_partition_count[4], list, ref[2][4]; | |
3349 | |
3350 if(h->slice_type == B_TYPE){ | |
3351 for(i=0; i<4; i++){ | |
3352 h->sub_mb_type[i]= get_ue_golomb(&s->gb); | |
3353 if(h->sub_mb_type[i] >=13){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3354 av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y); |
1168 | 3355 return -1; |
3356 } | |
3357 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; | |
3358 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type; | |
3359 } | |
3360 }else{ | |
3361 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ? | |
3362 for(i=0; i<4; i++){ | |
3363 h->sub_mb_type[i]= get_ue_golomb(&s->gb); | |
3364 if(h->sub_mb_type[i] >=4){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3365 av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y); |
1168 | 3366 return -1; |
3367 } | |
3368 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; | |
3369 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type; | |
3370 } | |
3371 } | |
3372 | |
3373 for(list=0; list<2; list++){ | |
3374 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list]; | |
3375 if(ref_count == 0) continue; | |
3376 for(i=0; i<4; i++){ | |
3377 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){ | |
3378 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip? | |
3379 }else{ | |
3380 //FIXME | |
3381 ref[list][i] = -1; | |
3382 } | |
3383 } | |
3384 } | |
3385 | |
3386 for(list=0; list<2; list++){ | |
3387 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list]; | |
3388 if(ref_count == 0) continue; | |
3389 | |
3390 for(i=0; i<4; i++){ | |
3391 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]= | |
3392 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i]; | |
3393 | |
3394 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){ | |
3395 const int sub_mb_type= h->sub_mb_type[i]; | |
3396 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1; | |
3397 for(j=0; j<sub_partition_count[i]; j++){ | |
3398 int mx, my; | |
3399 const int index= 4*i + block_width*j; | |
3400 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ]; | |
3401 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my); | |
3402 mx += get_se_golomb(&s->gb); | |
3403 my += get_se_golomb(&s->gb); | |
1170 | 3404 tprintf("final mv:%d %d\n", mx, my); |
3405 | |
1168 | 3406 if(IS_SUB_8X8(sub_mb_type)){ |
3407 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= | |
3408 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx; | |
3409 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= | |
3410 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my; | |
3411 }else if(IS_SUB_8X4(sub_mb_type)){ | |
3412 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx; | |
3413 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my; | |
3414 }else if(IS_SUB_4X8(sub_mb_type)){ | |
3415 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx; | |
3416 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my; | |
3417 }else{ | |
3418 assert(IS_SUB_4X4(sub_mb_type)); | |
3419 mv_cache[ 0 ][0]= mx; | |
3420 mv_cache[ 0 ][1]= my; | |
3421 } | |
3422 } | |
3423 }else{ | |
3424 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0]; | |
3425 p[0] = p[1]= | |
3426 p[8] = p[9]= 0; | |
3427 } | |
3428 } | |
3429 } | |
3430 }else if(!IS_DIRECT(mb_type)){ | |
3431 int list, mx, my, i; | |
3432 //FIXME we should set ref_idx_l? to 0 if we use that later ... | |
3433 if(IS_16X16(mb_type)){ | |
3434 for(list=0; list<2; list++){ | |
3435 if(h->ref_count[0]>0){ | |
3436 if(IS_DIR(mb_type, 0, list)){ | |
3437 const int val= get_te0_golomb(&s->gb, h->ref_count[list]); | |
3438 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1); | |
3439 } | |
3440 } | |
3441 } | |
3442 for(list=0; list<2; list++){ | |
3443 if(IS_DIR(mb_type, 0, list)){ | |
3444 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my); | |
3445 mx += get_se_golomb(&s->gb); | |
3446 my += get_se_golomb(&s->gb); | |
1170 | 3447 tprintf("final mv:%d %d\n", mx, my); |
3448 | |
1269 | 3449 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4); |
1168 | 3450 } |
3451 } | |
3452 } | |
3453 else if(IS_16X8(mb_type)){ | |
3454 for(list=0; list<2; list++){ | |
3455 if(h->ref_count[list]>0){ | |
3456 for(i=0; i<2; i++){ | |
3457 if(IS_DIR(mb_type, i, list)){ | |
3458 const int val= get_te0_golomb(&s->gb, h->ref_count[list]); | |
3459 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1); | |
3460 } | |
3461 } | |
3462 } | |
3463 } | |
3464 for(list=0; list<2; list++){ | |
3465 for(i=0; i<2; i++){ | |
3466 if(IS_DIR(mb_type, i, list)){ | |
3467 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my); | |
3468 mx += get_se_golomb(&s->gb); | |
3469 my += get_se_golomb(&s->gb); | |
1170 | 3470 tprintf("final mv:%d %d\n", mx, my); |
3471 | |
1269 | 3472 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4); |
1168 | 3473 } |
3474 } | |
3475 } | |
3476 }else{ | |
3477 assert(IS_8X16(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)){ //FIXME optimize | |
3482 const int val= get_te0_golomb(&s->gb, h->ref_count[list]); | |
3483 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 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_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*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] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4); |
1168 | 3497 } |
3498 } | |
3499 } | |
3500 } | |
3501 } | |
3502 | |
3503 if(IS_INTER(mb_type)) | |
3504 write_back_motion(h, mb_type); | |
3505 | |
3506 if(!IS_INTRA16x16(mb_type)){ | |
3507 cbp= get_ue_golomb(&s->gb); | |
3508 if(cbp > 47){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3509 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y); |
1168 | 3510 return -1; |
3511 } | |
3512 | |
3513 if(IS_INTRA4x4(mb_type)) | |
3514 cbp= golomb_to_intra4x4_cbp[cbp]; | |
3515 else | |
3516 cbp= golomb_to_inter_cbp[cbp]; | |
3517 } | |
3518 | |
3519 if(cbp || IS_INTRA16x16(mb_type)){ | |
3520 int i8x8, i4x4, chroma_idx; | |
3521 int chroma_qp, dquant; | |
3522 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr; | |
3523 const uint8_t *scan, *dc_scan; | |
3524 | |
3525 // fill_non_zero_count_cache(h); | |
3526 | |
3527 if(IS_INTERLACED(mb_type)){ | |
3528 scan= field_scan; | |
3529 dc_scan= luma_dc_field_scan; | |
3530 }else{ | |
3531 scan= zigzag_scan; | |
3532 dc_scan= luma_dc_zigzag_scan; | |
3533 } | |
3534 | |
3535 dquant= get_se_golomb(&s->gb); | |
3536 | |
3537 if( dquant > 25 || dquant < -26 ){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3538 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y); |
1168 | 3539 return -1; |
3540 } | |
3541 | |
3542 s->qscale += dquant; | |
3543 if(((unsigned)s->qscale) > 51){ | |
3544 if(s->qscale<0) s->qscale+= 52; | |
3545 else s->qscale-= 52; | |
3546 } | |
3547 | |
3548 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale); | |
3549 if(IS_INTRA16x16(mb_type)){ | |
3550 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){ | |
3551 return -1; //FIXME continue if partotioned and other retirn -1 too | |
3552 } | |
3553 | |
3554 assert((cbp&15) == 0 || (cbp&15) == 15); | |
3555 | |
3556 if(cbp&15){ | |
3557 for(i8x8=0; i8x8<4; i8x8++){ | |
3558 for(i4x4=0; i4x4<4; i4x4++){ | |
3559 const int index= i4x4 + 4*i8x8; | |
3560 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){ | |
3561 return -1; | |
3562 } | |
3563 } | |
3564 } | |
3565 }else{ | |
1636 | 3566 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1); |
1168 | 3567 } |
3568 }else{ | |
3569 for(i8x8=0; i8x8<4; i8x8++){ | |
3570 if(cbp & (1<<i8x8)){ | |
3571 for(i4x4=0; i4x4<4; i4x4++){ | |
3572 const int index= i4x4 + 4*i8x8; | |
3573 | |
3574 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){ | |
3575 return -1; | |
3576 } | |
3577 } | |
3578 }else{ | |
3579 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ]; | |
3580 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0; | |
3581 } | |
3582 } | |
3583 } | |
3584 | |
3585 if(cbp&0x30){ | |
3586 for(chroma_idx=0; chroma_idx<2; chroma_idx++) | |
3587 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){ | |
3588 return -1; | |
3589 } | |
3590 } | |
3591 | |
3592 if(cbp&0x20){ | |
3593 for(chroma_idx=0; chroma_idx<2; chroma_idx++){ | |
3594 for(i4x4=0; i4x4<4; i4x4++){ | |
3595 const int index= 16 + 4*chroma_idx + i4x4; | |
3596 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){ | |
3597 return -1; | |
3598 } | |
3599 } | |
3600 } | |
3601 }else{ | |
3602 uint8_t * const nnz= &h->non_zero_count_cache[0]; | |
3603 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] = | |
3604 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0; | |
3605 } | |
3606 }else{ | |
3607 memset(&h->non_zero_count_cache[8], 0, 8*5); | |
3608 } | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3609 s->current_picture.qscale_table[mb_xy]= s->qscale; |
1168 | 3610 write_back_non_zero_count(h); |
3611 | |
3612 return 0; | |
3613 } | |
3614 | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3615 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3616 int i, d; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3617 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3618 const int alpha = alpha_table[index_a]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3619 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3620 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3621 for( i = 0; i < 4; i++ ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3622 if( bS[i] == 0 ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3623 pix += 4 * stride; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3624 continue; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3625 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3626 |
1898 | 3627 if( bS[i] < 4 ) { |
3628 const int tc0 = tc0_table[index_a][bS[i] - 1]; | |
3629 /* 4px edge length */ | |
3630 for( d = 0; d < 4; d++ ) { | |
3631 const int p0 = pix[-1]; | |
3632 const int p1 = pix[-2]; | |
3633 const int p2 = pix[-3]; | |
3634 const int q0 = pix[0]; | |
3635 const int q1 = pix[1]; | |
3636 const int q2 = pix[2]; | |
3637 | |
3638 if( ABS( p0 - q0 ) < alpha && | |
3639 ABS( p1 - p0 ) < beta && | |
3640 ABS( q1 - q0 ) < beta ) { | |
3641 int tc = tc0; | |
3642 int i_delta; | |
3643 | |
3644 if( ABS( p2 - p0 ) < beta ) { | |
3645 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 ); | |
3646 tc++; | |
3647 } | |
3648 if( ABS( q2 - q0 ) < beta ) { | |
3649 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 ); | |
3650 tc++; | |
3651 } | |
3652 | |
3653 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); | |
3654 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */ | |
3655 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */ | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3656 } |
1898 | 3657 pix += stride; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3658 } |
1898 | 3659 }else{ |
3660 /* 4px edge length */ | |
3661 for( d = 0; d < 4; d++ ) { | |
3662 const int p0 = pix[-1]; | |
3663 const int p1 = pix[-2]; | |
3664 const int p2 = pix[-3]; | |
3665 | |
3666 const int q0 = pix[0]; | |
3667 const int q1 = pix[1]; | |
3668 const int q2 = pix[2]; | |
3669 | |
3670 if( ABS( p0 - q0 ) < alpha && | |
3671 ABS( p1 - p0 ) < beta && | |
3672 ABS( q1 - q0 ) < beta ) { | |
3673 | |
3674 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){ | |
3675 if( ABS( p2 - p0 ) < beta) | |
3676 { | |
3677 const int p3 = pix[-4]; | |
3678 /* p0', p1', p2' */ | |
3679 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3; | |
3680 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2; | |
3681 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3; | |
3682 } else { | |
3683 /* p0' */ | |
3684 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
3685 } | |
3686 if( ABS( q2 - q0 ) < beta) | |
3687 { | |
3688 const int q3 = pix[3]; | |
3689 /* q0', q1', q2' */ | |
3690 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3; | |
3691 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2; | |
3692 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3; | |
3693 } else { | |
3694 /* q0' */ | |
3695 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
3696 } | |
3697 }else{ | |
3698 /* p0', q0' */ | |
3699 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
3700 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
3701 } | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3702 } |
1898 | 3703 pix += stride; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3704 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3705 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3706 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3707 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3708 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3709 int i, d; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3710 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3711 const int alpha = alpha_table[index_a]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3712 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3713 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3714 for( i = 0; i < 4; i++ ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3715 if( bS[i] == 0 ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3716 pix += 2 * stride; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3717 continue; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3718 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3719 |
1898 | 3720 if( bS[i] < 4 ) { |
3721 const int tc = tc0_table[index_a][bS[i] - 1] + 1; | |
3722 /* 2px edge length (because we use same bS than the one for luma) */ | |
3723 for( d = 0; d < 2; d++ ){ | |
3724 const int p0 = pix[-1]; | |
3725 const int p1 = pix[-2]; | |
3726 const int q0 = pix[0]; | |
3727 const int q1 = pix[1]; | |
3728 | |
3729 if( ABS( p0 - q0 ) < alpha && | |
3730 ABS( p1 - p0 ) < beta && | |
3731 ABS( q1 - q0 ) < beta ) { | |
3732 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); | |
3733 | |
3734 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */ | |
3735 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */ | |
3736 } | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3737 pix += stride; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3738 } |
1898 | 3739 }else{ |
3740 /* 2px edge length (because we use same bS than the one for luma) */ | |
3741 for( d = 0; d < 2; d++ ){ | |
3742 const int p0 = pix[-1]; | |
3743 const int p1 = pix[-2]; | |
3744 const int q0 = pix[0]; | |
3745 const int q1 = pix[1]; | |
3746 | |
3747 if( ABS( p0 - q0 ) < alpha && | |
3748 ABS( p1 - p0 ) < beta && | |
3749 ABS( q1 - q0 ) < beta ) { | |
3750 | |
3751 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */ | |
3752 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */ | |
3753 } | |
3754 pix += stride; | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3755 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3756 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3757 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3758 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3759 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3760 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3761 int i, d; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3762 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3763 const int alpha = alpha_table[index_a]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3764 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3765 const int pix_next = stride; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3766 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3767 for( i = 0; i < 4; i++ ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3768 if( bS[i] == 0 ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3769 pix += 4; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3770 continue; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3771 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3772 |
1898 | 3773 if( bS[i] < 4 ) { |
3774 const int tc0 = tc0_table[index_a][bS[i] - 1]; | |
3775 /* 4px edge length */ | |
3776 for( d = 0; d < 4; d++ ) { | |
3777 const int p0 = pix[-1*pix_next]; | |
3778 const int p1 = pix[-2*pix_next]; | |
3779 const int p2 = pix[-3*pix_next]; | |
3780 const int q0 = pix[0]; | |
3781 const int q1 = pix[1*pix_next]; | |
3782 const int q2 = pix[2*pix_next]; | |
3783 | |
3784 if( ABS( p0 - q0 ) < alpha && | |
3785 ABS( p1 - p0 ) < beta && | |
3786 ABS( q1 - q0 ) < beta ) { | |
3787 | |
3788 int tc = tc0; | |
3789 int i_delta; | |
3790 | |
3791 if( ABS( p2 - p0 ) < beta ) { | |
3792 pix[-2*pix_next] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 ); | |
3793 tc++; | |
3794 } | |
3795 if( ABS( q2 - q0 ) < beta ) { | |
3796 pix[pix_next] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 ); | |
3797 tc++; | |
3798 } | |
3799 | |
3800 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); | |
3801 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */ | |
3802 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */ | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3803 } |
1898 | 3804 pix++; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3805 } |
1898 | 3806 }else{ |
3807 /* 4px edge length */ | |
3808 for( d = 0; d < 4; d++ ) { | |
3809 const int p0 = pix[-1*pix_next]; | |
3810 const int p1 = pix[-2*pix_next]; | |
3811 const int p2 = pix[-3*pix_next]; | |
3812 const int q0 = pix[0]; | |
3813 const int q1 = pix[1*pix_next]; | |
3814 const int q2 = pix[2*pix_next]; | |
3815 | |
3816 if( ABS( p0 - q0 ) < alpha && | |
3817 ABS( p1 - p0 ) < beta && | |
3818 ABS( q1 - q0 ) < beta ) { | |
3819 | |
3820 const int p3 = pix[-4*pix_next]; | |
3821 const int q3 = pix[ 3*pix_next]; | |
3822 | |
3823 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){ | |
3824 if( ABS( p2 - p0 ) < beta) { | |
3825 /* p0', p1', p2' */ | |
3826 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3; | |
3827 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2; | |
3828 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3; | |
3829 } else { | |
3830 /* p0' */ | |
3831 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
3832 } | |
3833 if( ABS( q2 - q0 ) < beta) { | |
3834 /* q0', q1', q2' */ | |
3835 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3; | |
3836 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2; | |
3837 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3; | |
3838 } else { | |
3839 /* q0' */ | |
3840 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
3841 } | |
3842 }else{ | |
3843 /* p0', q0' */ | |
3844 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
3845 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
3846 } | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3847 } |
1898 | 3848 pix++; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3849 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3850 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3851 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3852 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3853 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3854 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3855 int i, d; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3856 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3857 const int alpha = alpha_table[index_a]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3858 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3859 const int pix_next = stride; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3860 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3861 for( i = 0; i < 4; i++ ) |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3862 { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3863 if( bS[i] == 0 ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3864 pix += 2; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3865 continue; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3866 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3867 |
1898 | 3868 if( bS[i] < 4 ) { |
3869 int tc = tc0_table[index_a][bS[i] - 1] + 1; | |
3870 /* 2px edge length (see deblocking_filter_edgecv) */ | |
3871 for( d = 0; d < 2; d++ ) { | |
3872 const int p0 = pix[-1*pix_next]; | |
3873 const int p1 = pix[-2*pix_next]; | |
3874 const int q0 = pix[0]; | |
3875 const int q1 = pix[1*pix_next]; | |
3876 | |
3877 if( ABS( p0 - q0 ) < alpha && | |
3878 ABS( p1 - p0 ) < beta && | |
3879 ABS( q1 - q0 ) < beta ) { | |
3880 | |
3881 int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); | |
3882 | |
3883 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */ | |
3884 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */ | |
3885 } | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3886 pix++; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3887 } |
1898 | 3888 }else{ |
3889 /* 2px edge length (see deblocking_filter_edgecv) */ | |
3890 for( d = 0; d < 2; d++ ) { | |
3891 const int p0 = pix[-1*pix_next]; | |
3892 const int p1 = pix[-2*pix_next]; | |
3893 const int q0 = pix[0]; | |
3894 const int q1 = pix[1*pix_next]; | |
3895 | |
3896 if( ABS( p0 - q0 ) < alpha && | |
3897 ABS( p1 - p0 ) < beta && | |
3898 ABS( q1 - q0 ) < beta ) { | |
3899 | |
3900 pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */ | |
3901 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */ | |
3902 } | |
3903 pix++; | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3904 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3905 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3906 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3907 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3908 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3909 static void filter_mb( H264Context *h, int mb_x, int mb_y ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3910 MpegEncContext * const s = &h->s; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3911 const int mb_xy= mb_x + mb_y*s->mb_stride; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3912 uint8_t *img_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3913 uint8_t *img_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3914 uint8_t *img_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3915 int linesize, uvlinesize; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3916 int dir; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3917 #if 0 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3918 /* FIXME what's that ? */ |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3919 if( !s->decode ) |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3920 return; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3921 #endif |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3922 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3923 /* FIXME Implement deblocking filter for field MB */ |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3924 if( h->sps.mb_aff ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3925 return; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3926 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3927 linesize = s->linesize; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3928 uvlinesize = s->uvlinesize; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3929 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3930 /* dir : 0 -> vertical edge, 1 -> horizontal edge */ |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3931 for( dir = 0; dir < 2; dir++ ) |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3932 { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3933 int start = 0; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3934 int edge; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3935 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3936 /* test picture boundary */ |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3937 if( ( dir == 0 && mb_x == 0 ) || ( dir == 1 && mb_y == 0 ) ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3938 start = 1; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3939 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3940 /* FIXME test slice boundary */ |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3941 if( h->disable_deblocking_filter_idc == 2 ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3942 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3943 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3944 /* Calculate bS */ |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3945 for( edge = start; edge < 4; edge++ ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3946 /* mbn_xy: neighbour macroblock (how that works for field ?) */ |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3947 int mbn_xy = edge > 0 ? mb_xy : ( dir == 0 ? mb_xy -1 : mb_xy - s->mb_stride ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3948 int bS[4]; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3949 int qp; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3950 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3951 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) || |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3952 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3953 bS[0] = bS[1] = bS[2] = bS[3] = ( edge == 0 ? 4 : 3 ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3954 } else { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3955 int i; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3956 for( i = 0; i < 4; i++ ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3957 static const uint8_t block_idx_xy[4][4] = { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3958 { 0, 2, 8, 10}, { 1, 3, 9, 11}, |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3959 { 4, 6, 12, 14}, { 5, 7, 13, 15} |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3960 }; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3961 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3962 int x = dir == 0 ? edge : i; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3963 int y = dir == 0 ? i : edge; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3964 int xn = (x - (dir == 0 ? 1 : 0 ))&0x03; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3965 int yn = (y - (dir == 0 ? 0 : 1 ))&0x03; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3966 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3967 if( h->non_zero_count[mb_xy][block_idx_xy[x][y]] != 0 || |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3968 h->non_zero_count[mbn_xy][block_idx_xy[xn][yn]] != 0 ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3969 bS[i] = 2; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3970 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3971 else if( h->slice_type == P_TYPE ) { |
1898 | 3972 const int b8_xy = h->mb2b8_xy[mb_xy]+(y>>1)*h->b8_stride+(x>>1); |
3973 const int b8n_xy= h->mb2b8_xy[mbn_xy]+(yn>>1)*h->b8_stride+(xn>>1); | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3974 const int b_xy = h->mb2b_xy[mb_xy]+y*h->b_stride+x; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3975 const int bn_xy = h->mb2b_xy[mbn_xy]+yn*h->b_stride+xn; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3976 if( s->current_picture.ref_index[0][b8_xy] != s->current_picture.ref_index[0][b8n_xy] || |
1898 | 3977 ABS( s->current_picture.motion_val[0][b_xy][0] - s->current_picture.motion_val[0][bn_xy][0] ) >= 4 || |
3978 ABS( s->current_picture.motion_val[0][b_xy][1] - s->current_picture.motion_val[0][bn_xy][1] ) >= 4 ) | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3979 bS[i] = 1; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3980 else |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3981 bS[i] = 0; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3982 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3983 else { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3984 /* FIXME Add support for B frame */ |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3985 return; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3986 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3987 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3988 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3989 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3990 /* Filter edge */ |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3991 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3992 if( dir == 0 ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3993 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp ); |
1898 | 3994 if( (edge&1) == 0 ) { |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3995 int chroma_qp = ( get_chroma_qp( h, s->current_picture.qscale_table[mb_xy] ) + |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3996 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3997 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3998 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3999 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4000 } else { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4001 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp ); |
1898 | 4002 if( (edge&1) == 0 ) { |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4003 int chroma_qp = ( get_chroma_qp( h, s->current_picture.qscale_table[mb_xy] ) + |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4004 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4005 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4006 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4007 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4008 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4009 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4010 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4011 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4012 |
1168 | 4013 static int decode_slice(H264Context *h){ |
4014 MpegEncContext * const s = &h->s; | |
4015 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F; | |
4016 | |
4017 s->mb_skip_run= -1; | |
4018 | |
4019 #if 1 | |
4020 for(;;){ | |
4021 int ret= decode_mb(h); | |
4022 | |
4023 hl_decode_mb(h); | |
4024 | |
4025 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ? | |
4026 s->mb_y++; | |
4027 ret= decode_mb(h); | |
4028 | |
4029 hl_decode_mb(h); | |
4030 s->mb_y--; | |
4031 } | |
4032 | |
4033 if(ret<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4034 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y); |
1168 | 4035 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); |
4036 | |
4037 return -1; | |
4038 } | |
4039 | |
4040 if(++s->mb_x >= s->mb_width){ | |
4041 s->mb_x=0; | |
4042 ff_draw_horiz_band(s, 16*s->mb_y, 16); | |
4043 if(++s->mb_y >= s->mb_height){ | |
1170 | 4044 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits); |
4045 | |
1168 | 4046 if(get_bits_count(&s->gb) == s->gb.size_in_bits){ |
4047 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); | |
4048 | |
4049 return 0; | |
4050 }else{ | |
4051 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); | |
4052 | |
4053 return -1; | |
4054 } | |
4055 } | |
4056 } | |
4057 | |
4058 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){ | |
4059 if(get_bits_count(&s->gb) == s->gb.size_in_bits){ | |
4060 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); | |
4061 | |
4062 return 0; | |
4063 }else{ | |
4064 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); | |
4065 | |
4066 return -1; | |
4067 } | |
4068 } | |
4069 } | |
4070 #endif | |
4071 #if 0 | |
4072 for(;s->mb_y < s->mb_height; s->mb_y++){ | |
4073 for(;s->mb_x < s->mb_width; s->mb_x++){ | |
4074 int ret= decode_mb(h); | |
4075 | |
4076 hl_decode_mb(h); | |
4077 | |
4078 if(ret<0){ | |
4079 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y); | |
4080 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); | |
4081 | |
4082 return -1; | |
4083 } | |
4084 | |
4085 if(++s->mb_x >= s->mb_width){ | |
4086 s->mb_x=0; | |
4087 if(++s->mb_y >= s->mb_height){ | |
4088 if(get_bits_count(s->gb) == s->gb.size_in_bits){ | |
4089 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); | |
4090 | |
4091 return 0; | |
4092 }else{ | |
4093 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); | |
4094 | |
4095 return -1; | |
4096 } | |
4097 } | |
4098 } | |
4099 | |
4100 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){ | |
4101 if(get_bits_count(s->gb) == s->gb.size_in_bits){ | |
4102 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); | |
4103 | |
4104 return 0; | |
4105 }else{ | |
4106 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); | |
4107 | |
4108 return -1; | |
4109 } | |
4110 } | |
4111 } | |
4112 s->mb_x=0; | |
4113 ff_draw_horiz_band(s, 16*s->mb_y, 16); | |
4114 } | |
4115 #endif | |
4116 return -1; //not reached | |
4117 } | |
4118 | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4119 static void filter_frame(H264Context *h) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4120 int mb_x = 0; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4121 int mb_y = 0; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4122 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4123 for( mb_y = 0; mb_y < h->s.mb_height; mb_y++ ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4124 for( mb_x = 0; mb_x < h->s.mb_width; mb_x++ ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4125 filter_mb( h, mb_x, mb_y ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4126 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4127 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4128 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4129 |
1168 | 4130 static inline int decode_vui_parameters(H264Context *h, SPS *sps){ |
4131 MpegEncContext * const s = &h->s; | |
4132 int aspect_ratio_info_present_flag, aspect_ratio_idc; | |
4133 | |
4134 aspect_ratio_info_present_flag= get_bits1(&s->gb); | |
4135 | |
4136 if( aspect_ratio_info_present_flag ) { | |
4137 aspect_ratio_idc= get_bits(&s->gb, 8); | |
4138 if( aspect_ratio_idc == EXTENDED_SAR ) { | |
1548 | 4139 sps->sar.num= get_bits(&s->gb, 16); |
4140 sps->sar.den= get_bits(&s->gb, 16); | |
1168 | 4141 }else if(aspect_ratio_idc < 16){ |
1548 | 4142 sps->sar= pixel_aspect[aspect_ratio_idc]; |
1168 | 4143 }else{ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4144 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n"); |
1168 | 4145 return -1; |
4146 } | |
4147 }else{ | |
1548 | 4148 sps->sar.num= |
4149 sps->sar.den= 0; | |
1168 | 4150 } |
4151 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height); | |
4152 #if 0 | |
4153 | overscan_info_present_flag |0 |u(1) | | |
4154 | if( overscan_info_present_flag ) | | | | |
4155 | overscan_appropriate_flag |0 |u(1) | | |
4156 | video_signal_type_present_flag |0 |u(1) | | |
4157 | if( video_signal_type_present_flag ) { | | | | |
4158 | video_format |0 |u(3) | | |
4159 | video_full_range_flag |0 |u(1) | | |
4160 | colour_description_present_flag |0 |u(1) | | |
4161 | if( colour_description_present_flag ) { | | | | |
4162 | colour_primaries |0 |u(8) | | |
4163 | transfer_characteristics |0 |u(8) | | |
4164 | matrix_coefficients |0 |u(8) | | |
4165 | } | | | | |
4166 | } | | | | |
4167 | chroma_location_info_present_flag |0 |u(1) | | |
4168 | if ( chroma_location_info_present_flag ) { | | | | |
4169 | chroma_sample_location_type_top_field |0 |ue(v) | | |
4170 | chroma_sample_location_type_bottom_field |0 |ue(v) | | |
4171 | } | | | | |
4172 | timing_info_present_flag |0 |u(1) | | |
4173 | if( timing_info_present_flag ) { | | | | |
4174 | num_units_in_tick |0 |u(32) | | |
4175 | time_scale |0 |u(32) | | |
4176 | fixed_frame_rate_flag |0 |u(1) | | |
4177 | } | | | | |
4178 | nal_hrd_parameters_present_flag |0 |u(1) | | |
4179 | if( nal_hrd_parameters_present_flag = = 1) | | | | |
4180 | hrd_parameters( ) | | | | |
4181 | vcl_hrd_parameters_present_flag |0 |u(1) | | |
4182 | if( vcl_hrd_parameters_present_flag = = 1) | | | | |
4183 | hrd_parameters( ) | | | | |
4184 | if( ( nal_hrd_parameters_present_flag = = 1 | || | | | |
4185 | | | | | |
4186 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | | | |
4187 | low_delay_hrd_flag |0 |u(1) | | |
4188 | bitstream_restriction_flag |0 |u(1) | | |
4189 | if( bitstream_restriction_flag ) { |0 |u(1) | | |
4190 | motion_vectors_over_pic_boundaries_flag |0 |u(1) | | |
4191 | max_bytes_per_pic_denom |0 |ue(v) | | |
4192 | max_bits_per_mb_denom |0 |ue(v) | | |
4193 | log2_max_mv_length_horizontal |0 |ue(v) | | |
4194 | log2_max_mv_length_vertical |0 |ue(v) | | |
4195 | num_reorder_frames |0 |ue(v) | | |
4196 | max_dec_frame_buffering |0 |ue(v) | | |
4197 | } | | | | |
4198 |} | | | | |
4199 #endif | |
4200 return 0; | |
4201 } | |
4202 | |
4203 static inline int decode_seq_parameter_set(H264Context *h){ | |
4204 MpegEncContext * const s = &h->s; | |
1371 | 4205 int profile_idc, level_idc; |
1168 | 4206 int sps_id, i; |
4207 SPS *sps; | |
4208 | |
4209 profile_idc= get_bits(&s->gb, 8); | |
1371 | 4210 get_bits1(&s->gb); //constraint_set0_flag |
4211 get_bits1(&s->gb); //constraint_set1_flag | |
4212 get_bits1(&s->gb); //constraint_set2_flag | |
4213 get_bits(&s->gb, 5); // reserved | |
1168 | 4214 level_idc= get_bits(&s->gb, 8); |
4215 sps_id= get_ue_golomb(&s->gb); | |
4216 | |
4217 sps= &h->sps_buffer[ sps_id ]; | |
4218 sps->profile_idc= profile_idc; | |
4219 sps->level_idc= level_idc; | |
4220 | |
4221 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4; | |
4222 sps->poc_type= get_ue_golomb(&s->gb); | |
4223 | |
4224 if(sps->poc_type == 0){ //FIXME #define | |
4225 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4; | |
4226 } else if(sps->poc_type == 1){//FIXME #define | |
4227 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb); | |
4228 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb); | |
4229 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb); | |
4230 sps->poc_cycle_length= get_ue_golomb(&s->gb); | |
4231 | |
4232 for(i=0; i<sps->poc_cycle_length; i++) | |
4233 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb); | |
4234 } | |
4235 if(sps->poc_type > 2){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4236 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type); |
1168 | 4237 return -1; |
4238 } | |
4239 | |
4240 sps->ref_frame_count= get_ue_golomb(&s->gb); | |
1371 | 4241 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb); |
1168 | 4242 sps->mb_width= get_ue_golomb(&s->gb) + 1; |
4243 sps->mb_height= get_ue_golomb(&s->gb) + 1; | |
4244 sps->frame_mbs_only_flag= get_bits1(&s->gb); | |
4245 if(!sps->frame_mbs_only_flag) | |
4246 sps->mb_aff= get_bits1(&s->gb); | |
4247 else | |
4248 sps->mb_aff= 0; | |
4249 | |
4250 sps->direct_8x8_inference_flag= get_bits1(&s->gb); | |
4251 | |
1371 | 4252 sps->crop= get_bits1(&s->gb); |
4253 if(sps->crop){ | |
4254 sps->crop_left = get_ue_golomb(&s->gb); | |
4255 sps->crop_right = get_ue_golomb(&s->gb); | |
4256 sps->crop_top = get_ue_golomb(&s->gb); | |
4257 sps->crop_bottom= get_ue_golomb(&s->gb); | |
4258 if(sps->crop_left || sps->crop_top){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4259 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n"); |
1371 | 4260 } |
4261 }else{ | |
4262 sps->crop_left = | |
4263 sps->crop_right = | |
4264 sps->crop_top = | |
4265 sps->crop_bottom= 0; | |
4266 } | |
4267 | |
1168 | 4268 sps->vui_parameters_present_flag= get_bits1(&s->gb); |
4269 if( sps->vui_parameters_present_flag ) | |
4270 decode_vui_parameters(h, sps); | |
4271 | |
4272 if(s->avctx->debug&FF_DEBUG_PICT_INFO){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4273 av_log(h->s.avctx, AV_LOG_DEBUG, "sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s crop:%d/%d/%d/%d %s\n", |
1168 | 4274 sps_id, sps->profile_idc, sps->level_idc, |
4275 sps->poc_type, | |
4276 sps->ref_frame_count, | |
4277 sps->mb_width, sps->mb_height, | |
4278 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"), | |
4279 sps->direct_8x8_inference_flag ? "8B8" : "", | |
1371 | 4280 sps->crop_left, sps->crop_right, |
4281 sps->crop_top, sps->crop_bottom, | |
1168 | 4282 sps->vui_parameters_present_flag ? "VUI" : "" |
4283 ); | |
4284 } | |
4285 return 0; | |
4286 } | |
4287 | |
4288 static inline int decode_picture_parameter_set(H264Context *h){ | |
4289 MpegEncContext * const s = &h->s; | |
4290 int pps_id= get_ue_golomb(&s->gb); | |
4291 PPS *pps= &h->pps_buffer[pps_id]; | |
4292 | |
4293 pps->sps_id= get_ue_golomb(&s->gb); | |
4294 pps->cabac= get_bits1(&s->gb); | |
4295 pps->pic_order_present= get_bits1(&s->gb); | |
4296 pps->slice_group_count= get_ue_golomb(&s->gb) + 1; | |
4297 if(pps->slice_group_count > 1 ){ | |
4298 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb); | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4299 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n"); |
1168 | 4300 switch(pps->mb_slice_group_map_type){ |
4301 case 0: | |
4302 #if 0 | |
4303 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | | | |
4304 | run_length[ i ] |1 |ue(v) | | |
4305 #endif | |
4306 break; | |
4307 case 2: | |
4308 #if 0 | |
4309 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | | | |
4310 |{ | | | | |
4311 | top_left_mb[ i ] |1 |ue(v) | | |
4312 | bottom_right_mb[ i ] |1 |ue(v) | | |
4313 | } | | | | |
4314 #endif | |
4315 break; | |
4316 case 3: | |
4317 case 4: | |
4318 case 5: | |
4319 #if 0 | |
4320 | slice_group_change_direction_flag |1 |u(1) | | |
4321 | slice_group_change_rate_minus1 |1 |ue(v) | | |
4322 #endif | |
4323 break; | |
4324 case 6: | |
4325 #if 0 | |
4326 | slice_group_id_cnt_minus1 |1 |ue(v) | | |
4327 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | | | |
4328 |) | | | | |
4329 | slice_group_id[ i ] |1 |u(v) | | |
4330 #endif | |
1214 | 4331 break; |
1168 | 4332 } |
4333 } | |
4334 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1; | |
4335 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1; | |
4336 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4337 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n"); |
1168 | 4338 return -1; |
4339 } | |
4340 | |
4341 pps->weighted_pred= get_bits1(&s->gb); | |
4342 pps->weighted_bipred_idc= get_bits(&s->gb, 2); | |
4343 pps->init_qp= get_se_golomb(&s->gb) + 26; | |
4344 pps->init_qs= get_se_golomb(&s->gb) + 26; | |
4345 pps->chroma_qp_index_offset= get_se_golomb(&s->gb); | |
4346 pps->deblocking_filter_parameters_present= get_bits1(&s->gb); | |
4347 pps->constrained_intra_pred= get_bits1(&s->gb); | |
4348 pps->redundant_pic_cnt_present = get_bits1(&s->gb); | |
4349 | |
4350 if(s->avctx->debug&FF_DEBUG_PICT_INFO){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4351 av_log(h->s.avctx, AV_LOG_DEBUG, "pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s\n", |
1168 | 4352 pps_id, pps->sps_id, |
4353 pps->cabac ? "CABAC" : "CAVLC", | |
4354 pps->slice_group_count, | |
4355 pps->ref_count[0], pps->ref_count[1], | |
4356 pps->weighted_pred ? "weighted" : "", | |
4357 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset, | |
4358 pps->deblocking_filter_parameters_present ? "LPAR" : "", | |
4359 pps->constrained_intra_pred ? "CONSTR" : "", | |
1371 | 4360 pps->redundant_pic_cnt_present ? "REDU" : "" |
1168 | 4361 ); |
4362 } | |
4363 | |
4364 return 0; | |
4365 } | |
4366 | |
4367 /** | |
4368 * finds the end of the current frame in the bitstream. | |
4369 * @return the position of the first byte of the next frame, or -1 | |
4370 */ | |
4371 static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){ | |
4372 ParseContext *pc= &s->parse_context; | |
1187 | 4373 int i; |
1168 | 4374 uint32_t state; |
4375 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]); | |
4376 // mb_addr= pc->mb_addr - 1; | |
4377 state= pc->state; | |
4378 //FIXME this will fail with slices | |
4379 for(i=0; i<buf_size; i++){ | |
4380 state= (state<<8) | buf[i]; | |
4381 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){ | |
4382 if(pc->frame_start_found){ | |
4383 pc->state=-1; | |
4384 pc->frame_start_found= 0; | |
4385 return i-3; | |
4386 } | |
4387 pc->frame_start_found= 1; | |
4388 } | |
4389 } | |
4390 | |
4391 pc->state= state; | |
1219 | 4392 return END_NOT_FOUND; |
1168 | 4393 } |
4394 | |
4395 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){ | |
4396 MpegEncContext * const s = &h->s; | |
4397 AVCodecContext * const avctx= s->avctx; | |
4398 int buf_index=0; | |
1322 | 4399 #if 0 |
1168 | 4400 int i; |
4401 for(i=0; i<32; i++){ | |
4402 printf("%X ", buf[i]); | |
4403 } | |
4404 #endif | |
4405 for(;;){ | |
4406 int consumed; | |
4407 int dst_length; | |
4408 int bit_length; | |
4409 uint8_t *ptr; | |
4410 | |
4411 // start code prefix search | |
4412 for(; buf_index + 3 < buf_size; buf_index++){ | |
4413 // this should allways succeed in the first iteration | |
4414 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1) | |
4415 break; | |
4416 } | |
4417 | |
4418 if(buf_index+3 >= buf_size) break; | |
4419 | |
4420 buf_index+=3; | |
4421 | |
4422 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index); | |
4423 if(ptr[dst_length - 1] == 0) dst_length--; | |
4424 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1); | |
4425 | |
4426 if(s->avctx->debug&FF_DEBUG_STARTCODE){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4427 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length); |
1168 | 4428 } |
4429 | |
4430 buf_index += consumed; | |
4431 | |
4432 if(h->nal_ref_idc < s->hurry_up) | |
4433 continue; | |
4434 | |
4435 switch(h->nal_unit_type){ | |
4436 case NAL_IDR_SLICE: | |
4437 idr(h); //FIXME ensure we dont loose some frames if there is reordering | |
4438 case NAL_SLICE: | |
4439 init_get_bits(&s->gb, ptr, bit_length); | |
4440 h->intra_gb_ptr= | |
4441 h->inter_gb_ptr= &s->gb; | |
4442 s->data_partitioning = 0; | |
4443 | |
4444 if(decode_slice_header(h) < 0) return -1; | |
4445 if(h->redundant_pic_count==0) | |
4446 decode_slice(h); | |
4447 break; | |
4448 case NAL_DPA: | |
4449 init_get_bits(&s->gb, ptr, bit_length); | |
4450 h->intra_gb_ptr= | |
4451 h->inter_gb_ptr= NULL; | |
4452 s->data_partitioning = 1; | |
4453 | |
4454 if(decode_slice_header(h) < 0) return -1; | |
4455 break; | |
4456 case NAL_DPB: | |
4457 init_get_bits(&h->intra_gb, ptr, bit_length); | |
4458 h->intra_gb_ptr= &h->intra_gb; | |
4459 break; | |
4460 case NAL_DPC: | |
4461 init_get_bits(&h->inter_gb, ptr, bit_length); | |
4462 h->inter_gb_ptr= &h->inter_gb; | |
1174 | 4463 |
4464 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning) | |
1168 | 4465 decode_slice(h); |
4466 break; | |
4467 case NAL_SEI: | |
4468 break; | |
4469 case NAL_SPS: | |
4470 init_get_bits(&s->gb, ptr, bit_length); | |
4471 decode_seq_parameter_set(h); | |
4472 | |
4473 if(s->flags& CODEC_FLAG_LOW_DELAY) | |
4474 s->low_delay=1; | |
4475 | |
4476 avctx->has_b_frames= !s->low_delay; | |
4477 break; | |
4478 case NAL_PPS: | |
4479 init_get_bits(&s->gb, ptr, bit_length); | |
4480 | |
4481 decode_picture_parameter_set(h); | |
4482 | |
4483 break; | |
4484 case NAL_PICTURE_DELIMITER: | |
4485 break; | |
4486 case NAL_FILTER_DATA: | |
4487 break; | |
4488 } | |
4489 | |
4490 //FIXME move after where irt is set | |
4491 s->current_picture.pict_type= s->pict_type; | |
4492 s->current_picture.key_frame= s->pict_type == I_TYPE; | |
4493 } | |
4494 | |
1174 | 4495 if(!s->current_picture_ptr) return buf_index; //no frame |
4496 | |
1168 | 4497 h->prev_frame_num_offset= h->frame_num_offset; |
4498 h->prev_frame_num= h->frame_num; | |
4499 if(s->current_picture_ptr->reference){ | |
4500 h->prev_poc_msb= h->poc_msb; | |
4501 h->prev_poc_lsb= h->poc_lsb; | |
4502 } | |
4503 if(s->current_picture_ptr->reference) | |
4504 execute_ref_pic_marking(h, h->mmco, h->mmco_index); | |
4505 else | |
1169 | 4506 assert(h->mmco_index==0); |
1168 | 4507 |
4508 ff_er_frame_end(s); | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4509 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4510 if( h->disable_deblocking_filter_idc != 1 ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4511 filter_frame( h ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4512 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4513 |
1168 | 4514 MPV_frame_end(s); |
4515 | |
4516 return buf_index; | |
4517 } | |
4518 | |
4519 /** | |
4520 * retunrs the number of bytes consumed for building the current frame | |
4521 */ | |
4522 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){ | |
4523 if(s->flags&CODEC_FLAG_TRUNCATED){ | |
4524 pos -= s->parse_context.last_index; | |
4525 if(pos<0) pos=0; // FIXME remove (uneeded?) | |
4526 | |
4527 return pos; | |
4528 }else{ | |
4529 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...) | |
4530 if(pos+10>buf_size) pos=buf_size; // oops ;) | |
4531 | |
4532 return pos; | |
4533 } | |
4534 } | |
4535 | |
4536 static int decode_frame(AVCodecContext *avctx, | |
4537 void *data, int *data_size, | |
4538 uint8_t *buf, int buf_size) | |
4539 { | |
4540 H264Context *h = avctx->priv_data; | |
4541 MpegEncContext *s = &h->s; | |
4542 AVFrame *pict = data; | |
4543 int buf_index; | |
4544 | |
4545 s->flags= avctx->flags; | |
1754
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1706
diff
changeset
|
4546 s->flags2= avctx->flags2; |
1168 | 4547 |
4548 *data_size = 0; | |
4549 | |
4550 /* no supplementary picture */ | |
4551 if (buf_size == 0) { | |
4552 return 0; | |
4553 } | |
4554 | |
4555 if(s->flags&CODEC_FLAG_TRUNCATED){ | |
4556 int next= find_frame_end(s, buf, buf_size); | |
4557 | |
4558 if( ff_combine_frame(s, next, &buf, &buf_size) < 0 ) | |
4559 return buf_size; | |
4560 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index); | |
4561 } | |
4562 | |
4563 if(s->avctx->extradata_size && s->picture_number==0){ | |
4564 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) ) | |
4565 return -1; | |
4566 } | |
4567 | |
4568 buf_index=decode_nal_units(h, buf, buf_size); | |
4569 if(buf_index < 0) | |
4570 return -1; | |
4571 | |
4572 //FIXME do something with unavailable reference frames | |
4573 | |
4574 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size); | |
4575 #if 0 | |
4576 if(s->pict_type==B_TYPE || s->low_delay){ | |
4577 *pict= *(AVFrame*)&s->current_picture; | |
4578 } else { | |
4579 *pict= *(AVFrame*)&s->last_picture; | |
4580 } | |
4581 #endif | |
1174 | 4582 if(!s->current_picture_ptr){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4583 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n"); |
1174 | 4584 return -1; |
4585 } | |
4586 | |
1168 | 4587 *pict= *(AVFrame*)&s->current_picture; //FIXME |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1698
diff
changeset
|
4588 ff_print_debug_info(s, pict); |
1168 | 4589 assert(pict->data[0]); |
4590 //printf("out %d\n", (int)pict->data[0]); | |
4591 #if 0 //? | |
4592 | |
4593 /* Return the Picture timestamp as the frame number */ | |
4594 /* we substract 1 because it is added on utils.c */ | |
4595 avctx->frame_number = s->picture_number - 1; | |
4596 #endif | |
4597 #if 0 | |
4598 /* dont output the last pic after seeking */ | |
4599 if(s->last_picture_ptr || s->low_delay) | |
4600 //Note this isnt a issue as a IDR pic should flush teh buffers | |
4601 #endif | |
4602 *data_size = sizeof(AVFrame); | |
4603 return get_consumed_bytes(s, buf_index, buf_size); | |
4604 } | |
4605 #if 0 | |
4606 static inline void fill_mb_avail(H264Context *h){ | |
4607 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
|
4608 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 4609 |
4610 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
|
4611 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
|
4612 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
|
4613 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num; |
1168 | 4614 }else{ |
4615 h->mb_avail[0]= | |
4616 h->mb_avail[1]= | |
4617 h->mb_avail[2]= 0; | |
4618 } | |
4619 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num; | |
4620 h->mb_avail[4]= 1; //FIXME move out | |
4621 h->mb_avail[5]= 0; //FIXME move out | |
4622 } | |
4623 #endif | |
4624 | |
4625 #if 0 //selftest | |
4626 #define COUNT 8000 | |
4627 #define SIZE (COUNT*40) | |
4628 int main(){ | |
4629 int i; | |
4630 uint8_t temp[SIZE]; | |
4631 PutBitContext pb; | |
4632 GetBitContext gb; | |
4633 // int int_temp[10000]; | |
4634 DSPContext dsp; | |
4635 AVCodecContext avctx; | |
4636 | |
4637 dsputil_init(&dsp, &avctx); | |
4638 | |
1522
79dddc5cd990
removed the obsolete and unused parameters of init_put_bits
alex
parents:
1453
diff
changeset
|
4639 init_put_bits(&pb, temp, SIZE); |
1168 | 4640 printf("testing unsigned exp golomb\n"); |
4641 for(i=0; i<COUNT; i++){ | |
4642 START_TIMER | |
4643 set_ue_golomb(&pb, i); | |
4644 STOP_TIMER("set_ue_golomb"); | |
4645 } | |
4646 flush_put_bits(&pb); | |
4647 | |
4648 init_get_bits(&gb, temp, 8*SIZE); | |
4649 for(i=0; i<COUNT; i++){ | |
4650 int j, s; | |
4651 | |
4652 s= show_bits(&gb, 24); | |
4653 | |
4654 START_TIMER | |
4655 j= get_ue_golomb(&gb); | |
4656 if(j != i){ | |
4657 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s); | |
4658 // return -1; | |
4659 } | |
4660 STOP_TIMER("get_ue_golomb"); | |
4661 } | |
4662 | |
4663 | |
1524 | 4664 init_put_bits(&pb, temp, SIZE); |
1168 | 4665 printf("testing signed exp golomb\n"); |
4666 for(i=0; i<COUNT; i++){ | |
4667 START_TIMER | |
4668 set_se_golomb(&pb, i - COUNT/2); | |
4669 STOP_TIMER("set_se_golomb"); | |
4670 } | |
4671 flush_put_bits(&pb); | |
4672 | |
4673 init_get_bits(&gb, temp, 8*SIZE); | |
4674 for(i=0; i<COUNT; i++){ | |
4675 int j, s; | |
4676 | |
4677 s= show_bits(&gb, 24); | |
4678 | |
4679 START_TIMER | |
4680 j= get_se_golomb(&gb); | |
4681 if(j != i - COUNT/2){ | |
4682 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s); | |
4683 // return -1; | |
4684 } | |
4685 STOP_TIMER("get_se_golomb"); | |
4686 } | |
4687 | |
4688 printf("testing 4x4 (I)DCT\n"); | |
4689 | |
4690 DCTELEM block[16]; | |
4691 uint8_t src[16], ref[16]; | |
4692 uint64_t error= 0, max_error=0; | |
4693 | |
4694 for(i=0; i<COUNT; i++){ | |
4695 int j; | |
4696 // printf("%d %d %d\n", r1, r2, (r2-r1)*16); | |
4697 for(j=0; j<16; j++){ | |
4698 ref[j]= random()%255; | |
4699 src[j]= random()%255; | |
4700 } | |
4701 | |
4702 h264_diff_dct_c(block, src, ref, 4); | |
4703 | |
4704 //normalize | |
4705 for(j=0; j<16; j++){ | |
4706 // printf("%d ", block[j]); | |
4707 block[j]= block[j]*4; | |
4708 if(j&1) block[j]= (block[j]*4 + 2)/5; | |
4709 if(j&4) block[j]= (block[j]*4 + 2)/5; | |
4710 } | |
4711 // printf("\n"); | |
4712 | |
4713 h264_add_idct_c(ref, block, 4); | |
4714 /* for(j=0; j<16; j++){ | |
4715 printf("%d ", ref[j]); | |
4716 } | |
4717 printf("\n");*/ | |
4718 | |
4719 for(j=0; j<16; j++){ | |
4720 int diff= ABS(src[j] - ref[j]); | |
4721 | |
4722 error+= diff*diff; | |
4723 max_error= FFMAX(max_error, diff); | |
4724 } | |
4725 } | |
4726 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error ); | |
4727 #if 0 | |
4728 printf("testing quantizer\n"); | |
4729 for(qp=0; qp<52; qp++){ | |
4730 for(i=0; i<16; i++) | |
4731 src1_block[i]= src2_block[i]= random()%255; | |
4732 | |
4733 } | |
4734 #endif | |
4735 printf("Testing NAL layer\n"); | |
4736 | |
4737 uint8_t bitstream[COUNT]; | |
4738 uint8_t nal[COUNT*2]; | |
4739 H264Context h; | |
4740 memset(&h, 0, sizeof(H264Context)); | |
4741 | |
4742 for(i=0; i<COUNT; i++){ | |
4743 int zeros= i; | |
4744 int nal_length; | |
4745 int consumed; | |
4746 int out_length; | |
4747 uint8_t *out; | |
4748 int j; | |
4749 | |
4750 for(j=0; j<COUNT; j++){ | |
4751 bitstream[j]= (random() % 255) + 1; | |
4752 } | |
4753 | |
4754 for(j=0; j<zeros; j++){ | |
4755 int pos= random() % COUNT; | |
4756 while(bitstream[pos] == 0){ | |
4757 pos++; | |
4758 pos %= COUNT; | |
4759 } | |
4760 bitstream[pos]=0; | |
4761 } | |
4762 | |
4763 START_TIMER | |
4764 | |
4765 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2); | |
4766 if(nal_length<0){ | |
4767 printf("encoding failed\n"); | |
4768 return -1; | |
4769 } | |
4770 | |
4771 out= decode_nal(&h, nal, &out_length, &consumed, nal_length); | |
4772 | |
4773 STOP_TIMER("NAL") | |
4774 | |
4775 if(out_length != COUNT){ | |
4776 printf("incorrect length %d %d\n", out_length, COUNT); | |
4777 return -1; | |
4778 } | |
4779 | |
4780 if(consumed != nal_length){ | |
4781 printf("incorrect consumed length %d %d\n", nal_length, consumed); | |
4782 return -1; | |
4783 } | |
4784 | |
4785 if(memcmp(bitstream, out, COUNT)){ | |
4786 printf("missmatch\n"); | |
4787 return -1; | |
4788 } | |
4789 } | |
4790 | |
4791 printf("Testing RBSP\n"); | |
4792 | |
4793 | |
4794 return 0; | |
4795 } | |
4796 #endif | |
4797 | |
4798 | |
4799 static int decode_end(AVCodecContext *avctx) | |
4800 { | |
4801 H264Context *h = avctx->priv_data; | |
4802 MpegEncContext *s = &h->s; | |
4803 | |
4804 free_tables(h); //FIXME cleanup init stuff perhaps | |
4805 MPV_common_end(s); | |
4806 | |
4807 // memset(h, 0, sizeof(H264Context)); | |
4808 | |
4809 return 0; | |
4810 } | |
4811 | |
4812 | |
4813 AVCodec h264_decoder = { | |
4814 "h264", | |
4815 CODEC_TYPE_VIDEO, | |
4816 CODEC_ID_H264, | |
4817 sizeof(H264Context), | |
4818 decode_init, | |
4819 NULL, | |
4820 decode_end, | |
4821 decode_frame, | |
1214 | 4822 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED, |
1168 | 4823 }; |
4824 | |
1234 | 4825 #include "svq3.c" |