Mercurial > libavcodec.hg
annotate h264.c @ 1905:2761950695cc libavcodec
* some significant clean-up of the dv_decode_ac (it looks real
simple now, take a look for yourself).
* additional optimizations of the decoder. It runs at 55fps now
on my desktop and it used to be ~45fps.
author | romansh |
---|---|
date | Thu, 25 Mar 2004 01:46:40 +0000 |
parents | 379a18a7131f |
children | e20fd60b215c |
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; | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
165 uint8_t (*top_border)[16+2*8]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
166 uint8_t left_border[17+2*9]; |
1168 | 167 |
168 /** | |
169 * non zero coeff count cache. | |
170 * is 64 if not available. | |
171 */ | |
172 uint8_t non_zero_count_cache[6*8]; | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
173 uint8_t (*non_zero_count)[16]; |
1168 | 174 |
175 /** | |
176 * Motion vector cache. | |
177 */ | |
178 int16_t mv_cache[2][5*8][2]; | |
179 int8_t ref_cache[2][5*8]; | |
180 #define LIST_NOT_USED -1 //FIXME rename? | |
181 #define PART_NOT_AVAILABLE -2 | |
182 | |
183 /** | |
184 * is 1 if the specific list MV&references are set to 0,0,-2. | |
185 */ | |
186 int mv_cache_clean[2]; | |
187 | |
188 int block_offset[16+8]; | |
189 int chroma_subblock_offset[16]; //FIXME remove | |
190 | |
191 uint16_t *mb2b_xy; //FIXME are these 4 a good idea? | |
192 uint16_t *mb2b8_xy; | |
193 int b_stride; | |
194 int b8_stride; | |
195 | |
1234 | 196 int halfpel_flag; |
197 int thirdpel_flag; | |
198 | |
1319 | 199 int unknown_svq3_flag; |
200 int next_slice_index; | |
201 | |
1168 | 202 SPS sps_buffer[MAX_SPS_COUNT]; |
203 SPS sps; ///< current sps | |
204 | |
205 PPS pps_buffer[MAX_PPS_COUNT]; | |
206 /** | |
207 * current pps | |
208 */ | |
209 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that? | |
210 | |
211 int slice_num; | |
212 uint8_t *slice_table_base; | |
213 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1 | |
214 int slice_type; | |
215 int slice_type_fixed; | |
216 | |
217 //interlacing specific flags | |
218 int mb_field_decoding_flag; | |
219 | |
220 int sub_mb_type[4]; | |
221 | |
222 //POC stuff | |
223 int poc_lsb; | |
224 int poc_msb; | |
225 int delta_poc_bottom; | |
226 int delta_poc[2]; | |
227 int frame_num; | |
228 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0 | |
229 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0 | |
230 int frame_num_offset; ///< for POC type 2 | |
231 int prev_frame_num_offset; ///< for POC type 2 | |
232 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2 | |
233 | |
234 /** | |
235 * frame_num for frames or 2*frame_num for field pics. | |
236 */ | |
237 int curr_pic_num; | |
238 | |
239 /** | |
240 * max_frame_num or 2*max_frame_num for field pics. | |
241 */ | |
242 int max_pic_num; | |
243 | |
244 //Weighted pred stuff | |
245 int luma_log2_weight_denom; | |
246 int chroma_log2_weight_denom; | |
247 int luma_weight[2][16]; | |
248 int luma_offset[2][16]; | |
249 int chroma_weight[2][16][2]; | |
250 int chroma_offset[2][16][2]; | |
251 | |
252 //deblock | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
253 int deblocking_filter; ///< disable_deblocking_filter_idc with 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
|
254 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
|
255 int slice_beta_offset; |
1168 | 256 |
257 int redundant_pic_count; | |
258 | |
259 int direct_spatial_mv_pred; | |
260 | |
261 /** | |
262 * num_ref_idx_l0/1_active_minus1 + 1 | |
263 */ | |
264 int ref_count[2];// FIXME split for AFF | |
265 Picture *short_ref[16]; | |
266 Picture *long_ref[16]; | |
267 Picture default_ref_list[2][32]; | |
268 Picture ref_list[2][32]; //FIXME size? | |
269 Picture field_ref_list[2][32]; //FIXME size? | |
270 | |
271 /** | |
272 * memory management control operations buffer. | |
273 */ | |
274 MMCO mmco[MAX_MMCO_COUNT]; | |
275 int mmco_index; | |
276 | |
277 int long_ref_count; ///< number of actual long term references | |
278 int short_ref_count; ///< number of actual short term references | |
279 | |
280 //data partitioning | |
281 GetBitContext intra_gb; | |
282 GetBitContext inter_gb; | |
283 GetBitContext *intra_gb_ptr; | |
284 GetBitContext *inter_gb_ptr; | |
285 | |
286 DCTELEM mb[16*24] __align8; | |
287 }H264Context; | |
288 | |
289 static VLC coeff_token_vlc[4]; | |
290 static VLC chroma_dc_coeff_token_vlc; | |
291 | |
292 static VLC total_zeros_vlc[15]; | |
293 static VLC chroma_dc_total_zeros_vlc[3]; | |
294 | |
295 static VLC run_vlc[6]; | |
296 static VLC run7_vlc; | |
297 | |
1234 | 298 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp); |
299 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc); | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
300 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr); |
1234 | 301 |
1269 | 302 static inline uint32_t pack16to32(int a, int b){ |
303 #ifdef WORDS_BIGENDIAN | |
304 return (b&0xFFFF) + (a<<16); | |
305 #else | |
306 return (a&0xFFFF) + (b<<16); | |
307 #endif | |
308 } | |
309 | |
1168 | 310 /** |
311 * fill a rectangle. | |
312 * @param h height of the recatangle, should be a constant | |
313 * @param w width of the recatangle, should be a constant | |
314 * @param size the size of val (1 or 4), should be a constant | |
315 */ | |
1187 | 316 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined |
317 uint8_t *p= (uint8_t*)vp; | |
1168 | 318 assert(size==1 || size==4); |
319 | |
320 w *= size; | |
321 stride *= size; | |
322 | |
323 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it | |
324 if(w==2 && h==2){ | |
325 *(uint16_t*)(p + 0)= | |
326 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101; | |
327 }else if(w==2 && h==4){ | |
328 *(uint16_t*)(p + 0*stride)= | |
329 *(uint16_t*)(p + 1*stride)= | |
330 *(uint16_t*)(p + 2*stride)= | |
331 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101; | |
1252 | 332 }else if(w==4 && h==1){ |
333 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101; | |
1168 | 334 }else if(w==4 && h==2){ |
335 *(uint32_t*)(p + 0*stride)= | |
336 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101; | |
337 }else if(w==4 && h==4){ | |
338 *(uint32_t*)(p + 0*stride)= | |
339 *(uint32_t*)(p + 1*stride)= | |
340 *(uint32_t*)(p + 2*stride)= | |
341 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101; | |
342 }else if(w==8 && h==1){ | |
343 *(uint32_t*)(p + 0)= | |
344 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101; | |
345 }else if(w==8 && h==2){ | |
346 *(uint32_t*)(p + 0 + 0*stride)= | |
347 *(uint32_t*)(p + 4 + 0*stride)= | |
348 *(uint32_t*)(p + 0 + 1*stride)= | |
349 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101; | |
350 }else if(w==8 && h==4){ | |
351 *(uint64_t*)(p + 0*stride)= | |
352 *(uint64_t*)(p + 1*stride)= | |
353 *(uint64_t*)(p + 2*stride)= | |
354 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL; | |
355 }else if(w==16 && h==2){ | |
356 *(uint64_t*)(p + 0+0*stride)= | |
357 *(uint64_t*)(p + 8+0*stride)= | |
358 *(uint64_t*)(p + 0+1*stride)= | |
359 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL; | |
360 }else if(w==16 && h==4){ | |
361 *(uint64_t*)(p + 0+0*stride)= | |
362 *(uint64_t*)(p + 8+0*stride)= | |
363 *(uint64_t*)(p + 0+1*stride)= | |
364 *(uint64_t*)(p + 8+1*stride)= | |
365 *(uint64_t*)(p + 0+2*stride)= | |
366 *(uint64_t*)(p + 8+2*stride)= | |
367 *(uint64_t*)(p + 0+3*stride)= | |
368 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL; | |
369 }else | |
370 assert(0); | |
371 } | |
372 | |
373 static inline void fill_caches(H264Context *h, int mb_type){ | |
374 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
|
375 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 376 int topleft_xy, top_xy, topright_xy, left_xy[2]; |
377 int topleft_type, top_type, topright_type, left_type[2]; | |
378 int left_block[4]; | |
379 int i; | |
380 | |
381 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it | |
382 | |
383 if(h->sps.mb_aff){ | |
384 //FIXME | |
1453 | 385 topleft_xy = 0; /* avoid warning */ |
386 top_xy = 0; /* avoid warning */ | |
387 topright_xy = 0; /* avoid warning */ | |
1168 | 388 }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
|
389 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
|
390 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
|
391 topright_xy= mb_xy+1 - s->mb_stride; |
1168 | 392 left_xy[0] = mb_xy-1; |
393 left_xy[1] = mb_xy-1; | |
394 left_block[0]= 0; | |
395 left_block[1]= 1; | |
396 left_block[2]= 2; | |
397 left_block[3]= 3; | |
398 } | |
399 | |
400 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0; | |
401 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0; | |
402 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0; | |
403 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0; | |
404 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0; | |
405 | |
406 if(IS_INTRA(mb_type)){ | |
407 h->topleft_samples_available= | |
408 h->top_samples_available= | |
409 h->left_samples_available= 0xFFFF; | |
410 h->topright_samples_available= 0xEEEA; | |
411 | |
412 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){ | |
413 h->topleft_samples_available= 0xB3FF; | |
414 h->top_samples_available= 0x33FF; | |
415 h->topright_samples_available= 0x26EA; | |
416 } | |
417 for(i=0; i<2; i++){ | |
418 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){ | |
419 h->topleft_samples_available&= 0xDF5F; | |
420 h->left_samples_available&= 0x5F5F; | |
421 } | |
422 } | |
423 | |
424 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred)) | |
425 h->topleft_samples_available&= 0x7FFF; | |
426 | |
427 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred)) | |
428 h->topright_samples_available&= 0xFBFF; | |
429 | |
430 if(IS_INTRA4x4(mb_type)){ | |
431 if(IS_INTRA4x4(top_type)){ | |
432 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4]; | |
433 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5]; | |
434 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6]; | |
435 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3]; | |
436 }else{ | |
437 int pred; | |
438 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred)) | |
439 pred= 2; | |
440 else{ | |
441 pred= -1; | |
442 } | |
443 h->intra4x4_pred_mode_cache[4+8*0]= | |
444 h->intra4x4_pred_mode_cache[5+8*0]= | |
445 h->intra4x4_pred_mode_cache[6+8*0]= | |
446 h->intra4x4_pred_mode_cache[7+8*0]= pred; | |
447 } | |
448 for(i=0; i<2; i++){ | |
449 if(IS_INTRA4x4(left_type[i])){ | |
450 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]]; | |
451 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]]; | |
452 }else{ | |
453 int pred; | |
454 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred)) | |
455 pred= 2; | |
456 else{ | |
457 pred= -1; | |
458 } | |
459 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= | |
460 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred; | |
461 } | |
462 } | |
463 } | |
464 } | |
465 | |
466 | |
467 /* | |
468 0 . T T. T T T T | |
469 1 L . .L . . . . | |
470 2 L . .L . . . . | |
471 3 . T TL . . . . | |
472 4 L . .L . . . . | |
473 5 L . .. . . . . | |
474 */ | |
475 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec) | |
476 if(top_type){ | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
477 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
478 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
479 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
480 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3]; |
1168 | 481 |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
482 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
483 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8]; |
1168 | 484 |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
485 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
486 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11]; |
1168 | 487 }else{ |
488 h->non_zero_count_cache[4+8*0]= | |
489 h->non_zero_count_cache[5+8*0]= | |
490 h->non_zero_count_cache[6+8*0]= | |
491 h->non_zero_count_cache[7+8*0]= | |
492 | |
493 h->non_zero_count_cache[1+8*0]= | |
494 h->non_zero_count_cache[2+8*0]= | |
495 | |
496 h->non_zero_count_cache[1+8*3]= | |
497 h->non_zero_count_cache[2+8*3]= 64; | |
498 } | |
499 | |
500 if(left_type[0]){ | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
501 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
502 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
503 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
504 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12]; |
1168 | 505 }else{ |
506 h->non_zero_count_cache[3+8*1]= | |
507 h->non_zero_count_cache[3+8*2]= | |
508 h->non_zero_count_cache[0+8*1]= | |
509 h->non_zero_count_cache[0+8*4]= 64; | |
510 } | |
511 | |
512 if(left_type[1]){ | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
513 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
514 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
515 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
516 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11]; |
1168 | 517 }else{ |
518 h->non_zero_count_cache[3+8*3]= | |
519 h->non_zero_count_cache[3+8*4]= | |
520 h->non_zero_count_cache[0+8*2]= | |
521 h->non_zero_count_cache[0+8*5]= 64; | |
522 } | |
523 | |
524 #if 1 | |
525 if(IS_INTER(mb_type)){ | |
526 int list; | |
527 for(list=0; list<2; list++){ | |
528 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){ | |
529 /*if(!h->mv_cache_clean[list]){ | |
530 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all? | |
531 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t)); | |
532 h->mv_cache_clean[list]= 1; | |
533 }*/ | |
534 continue; //FIXME direct mode ... | |
535 } | |
536 h->mv_cache_clean[list]= 0; | |
537 | |
538 if(IS_INTER(topleft_type)){ | |
539 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride; | |
540 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride; | |
541 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy]; | |
542 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy]; | |
543 }else{ | |
544 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0; | |
545 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
546 } | |
547 | |
548 if(IS_INTER(top_type)){ | |
549 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride; | |
550 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride; | |
551 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0]; | |
552 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1]; | |
553 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2]; | |
554 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3]; | |
555 h->ref_cache[list][scan8[0] + 0 - 1*8]= | |
556 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0]; | |
557 h->ref_cache[list][scan8[0] + 2 - 1*8]= | |
558 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1]; | |
559 }else{ | |
560 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= | |
561 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= | |
562 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= | |
563 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0; | |
564 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101; | |
565 } | |
566 | |
567 if(IS_INTER(topright_type)){ | |
568 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride; | |
569 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride; | |
570 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy]; | |
571 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy]; | |
572 }else{ | |
573 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0; | |
574 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
575 } | |
576 | |
577 //FIXME unify cleanup or sth | |
578 if(IS_INTER(left_type[0])){ | |
579 const int b_xy= h->mb2b_xy[left_xy[0]] + 3; | |
580 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1; | |
581 *(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]]; | |
582 *(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]]; | |
583 h->ref_cache[list][scan8[0] - 1 + 0*8]= | |
584 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)]; | |
585 }else{ | |
586 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]= | |
587 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0; | |
588 h->ref_cache[list][scan8[0] - 1 + 0*8]= | |
589 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
590 } | |
591 | |
592 if(IS_INTER(left_type[1])){ | |
593 const int b_xy= h->mb2b_xy[left_xy[1]] + 3; | |
594 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1; | |
595 *(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]]; | |
596 *(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]]; | |
597 h->ref_cache[list][scan8[0] - 1 + 2*8]= | |
598 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)]; | |
599 }else{ | |
600 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]= | |
601 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0; | |
602 h->ref_cache[list][scan8[0] - 1 + 2*8]= | |
603 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
604 } | |
605 | |
606 h->ref_cache[list][scan8[5 ]+1] = | |
607 h->ref_cache[list][scan8[7 ]+1] = | |
608 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else) | |
609 h->ref_cache[list][scan8[4 ]] = | |
610 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE; | |
611 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]= | |
612 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]= | |
613 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else) | |
614 *(uint32_t*)h->mv_cache [list][scan8[4 ]]= | |
615 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0; | |
616 } | |
617 //FIXME | |
618 | |
619 } | |
620 #endif | |
621 } | |
622 | |
623 static inline void write_back_intra_pred_mode(H264Context *h){ | |
624 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
|
625 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 626 |
627 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1]; | |
628 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2]; | |
629 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3]; | |
630 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4]; | |
631 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4]; | |
632 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4]; | |
633 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4]; | |
634 } | |
635 | |
636 /** | |
637 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. | |
638 */ | |
639 static inline int check_intra4x4_pred_mode(H264Context *h){ | |
640 MpegEncContext * const s = &h->s; | |
641 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0}; | |
642 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED}; | |
643 int i; | |
644 | |
645 if(!(h->top_samples_available&0x8000)){ | |
646 for(i=0; i<4; i++){ | |
647 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ]; | |
648 if(status<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
649 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 | 650 return -1; |
651 } else if(status){ | |
652 h->intra4x4_pred_mode_cache[scan8[0] + i]= status; | |
653 } | |
654 } | |
655 } | |
656 | |
657 if(!(h->left_samples_available&0x8000)){ | |
658 for(i=0; i<4; i++){ | |
659 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ]; | |
660 if(status<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
661 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 | 662 return -1; |
663 } else if(status){ | |
664 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status; | |
665 } | |
666 } | |
667 } | |
668 | |
669 return 0; | |
670 } //FIXME cleanup like next | |
671 | |
672 /** | |
673 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. | |
674 */ | |
675 static inline int check_intra_pred_mode(H264Context *h, int mode){ | |
676 MpegEncContext * const s = &h->s; | |
677 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1}; | |
678 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8}; | |
679 | |
680 if(!(h->top_samples_available&0x8000)){ | |
681 mode= top[ mode ]; | |
682 if(mode<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
683 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 | 684 return -1; |
685 } | |
686 } | |
687 | |
688 if(!(h->left_samples_available&0x8000)){ | |
689 mode= left[ mode ]; | |
690 if(mode<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
691 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 | 692 return -1; |
693 } | |
694 } | |
695 | |
696 return mode; | |
697 } | |
698 | |
699 /** | |
700 * gets the predicted intra4x4 prediction mode. | |
701 */ | |
702 static inline int pred_intra_mode(H264Context *h, int n){ | |
703 const int index8= scan8[n]; | |
704 const int left= h->intra4x4_pred_mode_cache[index8 - 1]; | |
705 const int top = h->intra4x4_pred_mode_cache[index8 - 8]; | |
706 const int min= FFMIN(left, top); | |
707 | |
1170 | 708 tprintf("mode:%d %d min:%d\n", left ,top, min); |
1168 | 709 |
710 if(min<0) return DC_PRED; | |
711 else return min; | |
712 } | |
713 | |
714 static inline void write_back_non_zero_count(H264Context *h){ | |
715 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
|
716 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
717 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
718 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
719 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
720 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
721 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
722 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
723 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
724 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
725 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
726 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
727 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
728 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
729 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
730 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
731 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
732 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4]; |
1168 | 733 } |
734 | |
735 /** | |
736 * gets the predicted number of non zero coefficients. | |
737 * @param n block index | |
738 */ | |
739 static inline int pred_non_zero_count(H264Context *h, int n){ | |
740 const int index8= scan8[n]; | |
741 const int left= h->non_zero_count_cache[index8 - 1]; | |
742 const int top = h->non_zero_count_cache[index8 - 8]; | |
743 int i= left + top; | |
744 | |
745 if(i<64) i= (i+1)>>1; | |
746 | |
1170 | 747 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31); |
1168 | 748 |
749 return i&31; | |
750 } | |
751 | |
1169 | 752 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){ |
753 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ]; | |
754 | |
755 if(topright_ref != PART_NOT_AVAILABLE){ | |
756 *C= h->mv_cache[list][ i - 8 + part_width ]; | |
757 return topright_ref; | |
758 }else{ | |
1170 | 759 tprintf("topright MV not available\n"); |
760 | |
1169 | 761 *C= h->mv_cache[list][ i - 8 - 1 ]; |
762 return h->ref_cache[list][ i - 8 - 1 ]; | |
763 } | |
764 } | |
765 | |
1168 | 766 /** |
767 * gets the predicted MV. | |
768 * @param n the block index | |
769 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4) | |
770 * @param mx the x component of the predicted motion vector | |
771 * @param my the y component of the predicted motion vector | |
772 */ | |
773 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){ | |
774 const int index8= scan8[n]; | |
775 const int top_ref= h->ref_cache[list][ index8 - 8 ]; | |
776 const int left_ref= h->ref_cache[list][ index8 - 1 ]; | |
777 const int16_t * const A= h->mv_cache[list][ index8 - 1 ]; | |
778 const int16_t * const B= h->mv_cache[list][ index8 - 8 ]; | |
1169 | 779 const int16_t * C; |
780 int diagonal_ref, match_count; | |
781 | |
1168 | 782 assert(part_width==1 || part_width==2 || part_width==4); |
1169 | 783 |
1168 | 784 /* mv_cache |
785 B . . A T T T T | |
786 U . . L . . , . | |
787 U . . L . . . . | |
788 U . . L . . , . | |
789 . . . L . . . . | |
790 */ | |
1169 | 791 |
792 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width); | |
793 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref); | |
794 if(match_count > 1){ //most common | |
795 *mx= mid_pred(A[0], B[0], C[0]); | |
796 *my= mid_pred(A[1], B[1], C[1]); | |
797 }else if(match_count==1){ | |
798 if(left_ref==ref){ | |
799 *mx= A[0]; | |
800 *my= A[1]; | |
801 }else if(top_ref==ref){ | |
802 *mx= B[0]; | |
803 *my= B[1]; | |
804 }else{ | |
805 *mx= C[0]; | |
806 *my= C[1]; | |
807 } | |
808 }else{ | |
809 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){ | |
810 *mx= A[0]; | |
811 *my= A[1]; | |
1168 | 812 }else{ |
813 *mx= mid_pred(A[0], B[0], C[0]); | |
814 *my= mid_pred(A[1], B[1], C[1]); | |
815 } | |
1169 | 816 } |
1168 | 817 |
1187 | 818 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 | 819 } |
820 | |
821 /** | |
822 * gets the directionally predicted 16x8 MV. | |
823 * @param n the block index | |
824 * @param mx the x component of the predicted motion vector | |
825 * @param my the y component of the predicted motion vector | |
826 */ | |
827 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){ | |
828 if(n==0){ | |
829 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ]; | |
830 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ]; | |
831 | |
1187 | 832 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 | 833 |
834 if(top_ref == ref){ | |
835 *mx= B[0]; | |
836 *my= B[1]; | |
837 return; | |
838 } | |
839 }else{ | |
840 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ]; | |
841 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ]; | |
842 | |
1187 | 843 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 | 844 |
845 if(left_ref == ref){ | |
846 *mx= A[0]; | |
847 *my= A[1]; | |
848 return; | |
849 } | |
850 } | |
851 | |
852 //RARE | |
853 pred_motion(h, n, 4, list, ref, mx, my); | |
854 } | |
855 | |
856 /** | |
857 * gets the directionally predicted 8x16 MV. | |
858 * @param n the block index | |
859 * @param mx the x component of the predicted motion vector | |
860 * @param my the y component of the predicted motion vector | |
861 */ | |
862 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){ | |
863 if(n==0){ | |
864 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ]; | |
865 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ]; | |
866 | |
1187 | 867 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 | 868 |
869 if(left_ref == ref){ | |
870 *mx= A[0]; | |
871 *my= A[1]; | |
872 return; | |
873 } | |
874 }else{ | |
1169 | 875 const int16_t * C; |
876 int diagonal_ref; | |
877 | |
878 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2); | |
1168 | 879 |
1187 | 880 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 | 881 |
1169 | 882 if(diagonal_ref == ref){ |
1168 | 883 *mx= C[0]; |
884 *my= C[1]; | |
885 return; | |
886 } | |
887 } | |
888 | |
889 //RARE | |
890 pred_motion(h, n, 2, list, ref, mx, my); | |
891 } | |
892 | |
893 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){ | |
894 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ]; | |
895 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ]; | |
896 | |
1187 | 897 tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y); |
1168 | 898 |
899 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE | |
900 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0) | |
901 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){ | |
902 | |
903 *mx = *my = 0; | |
904 return; | |
905 } | |
906 | |
907 pred_motion(h, 0, 4, 0, 0, mx, my); | |
908 | |
909 return; | |
910 } | |
911 | |
912 static inline void write_back_motion(H264Context *h, int mb_type){ | |
913 MpegEncContext * const s = &h->s; | |
914 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; | |
915 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride; | |
916 int list; | |
917 | |
918 for(list=0; list<2; list++){ | |
919 int y; | |
920 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){ | |
921 if(1){ //FIXME skip or never read if mb_type doesnt use it | |
922 for(y=0; y<4; y++){ | |
923 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= | |
924 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0; | |
925 } | |
926 for(y=0; y<2; y++){ | |
927 *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101; | |
928 } | |
929 } | |
930 continue; //FIXME direct mode ... | |
931 } | |
932 | |
933 for(y=0; y<4; y++){ | |
934 *(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]; | |
935 *(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]; | |
936 } | |
937 for(y=0; y<2; y++){ | |
938 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y]; | |
939 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y]; | |
940 } | |
941 } | |
942 } | |
943 | |
944 /** | |
945 * Decodes a network abstraction layer unit. | |
946 * @param consumed is the number of bytes used as input | |
947 * @param length is the length of the array | |
948 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing? | |
949 * @returns decoded bytes, might be src+1 if no escapes | |
950 */ | |
951 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){ | |
952 int i, si, di; | |
953 uint8_t *dst; | |
954 | |
955 // src[0]&0x80; //forbidden bit | |
956 h->nal_ref_idc= src[0]>>5; | |
957 h->nal_unit_type= src[0]&0x1F; | |
958 | |
959 src++; length--; | |
960 #if 0 | |
961 for(i=0; i<length; i++) | |
962 printf("%2X ", src[i]); | |
963 #endif | |
964 for(i=0; i+1<length; i+=2){ | |
965 if(src[i]) continue; | |
966 if(i>0 && src[i-1]==0) i--; | |
967 if(i+2<length && src[i+1]==0 && src[i+2]<=3){ | |
968 if(src[i+2]!=3){ | |
969 /* startcode, so we must be past the end */ | |
970 length=i; | |
971 } | |
972 break; | |
973 } | |
974 } | |
975 | |
976 if(i>=length-1){ //no escaped 0 | |
977 *dst_length= length; | |
978 *consumed= length+1; //+1 for the header | |
979 return src; | |
980 } | |
981 | |
982 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length); | |
983 dst= h->rbsp_buffer; | |
984 | |
985 //printf("deoding esc\n"); | |
986 si=di=0; | |
987 while(si<length){ | |
988 //remove escapes (very rare 1:2^22) | |
989 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){ | |
990 if(src[si+2]==3){ //escape | |
991 dst[di++]= 0; | |
992 dst[di++]= 0; | |
993 si+=3; | |
994 }else //next start code | |
995 break; | |
996 } | |
997 | |
998 dst[di++]= src[si++]; | |
999 } | |
1000 | |
1001 *dst_length= di; | |
1002 *consumed= si + 1;//+1 for the header | |
1003 //FIXME store exact number of bits in the getbitcontext (its needed for decoding) | |
1004 return dst; | |
1005 } | |
1006 | |
1007 /** | |
1008 * @param src the data which should be escaped | |
1009 * @param dst the target buffer, dst+1 == src is allowed as a special case | |
1010 * @param length the length of the src data | |
1011 * @param dst_length the length of the dst array | |
1012 * @returns length of escaped data in bytes or -1 if an error occured | |
1013 */ | |
1014 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){ | |
1015 int i, escape_count, si, di; | |
1016 uint8_t *temp; | |
1017 | |
1018 assert(length>=0); | |
1019 assert(dst_length>0); | |
1020 | |
1021 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type; | |
1022 | |
1023 if(length==0) return 1; | |
1024 | |
1025 escape_count= 0; | |
1026 for(i=0; i<length; i+=2){ | |
1027 if(src[i]) continue; | |
1028 if(i>0 && src[i-1]==0) | |
1029 i--; | |
1030 if(i+2<length && src[i+1]==0 && src[i+2]<=3){ | |
1031 escape_count++; | |
1032 i+=2; | |
1033 } | |
1034 } | |
1035 | |
1036 if(escape_count==0){ | |
1037 if(dst+1 != src) | |
1038 memcpy(dst+1, src, length); | |
1039 return length + 1; | |
1040 } | |
1041 | |
1042 if(length + escape_count + 1> dst_length) | |
1043 return -1; | |
1044 | |
1045 //this should be damn rare (hopefully) | |
1046 | |
1047 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count); | |
1048 temp= h->rbsp_buffer; | |
1049 //printf("encoding esc\n"); | |
1050 | |
1051 si= 0; | |
1052 di= 0; | |
1053 while(si < length){ | |
1054 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){ | |
1055 temp[di++]= 0; si++; | |
1056 temp[di++]= 0; si++; | |
1057 temp[di++]= 3; | |
1058 temp[di++]= src[si++]; | |
1059 } | |
1060 else | |
1061 temp[di++]= src[si++]; | |
1062 } | |
1063 memcpy(dst+1, temp, length+escape_count); | |
1064 | |
1065 assert(di == length+escape_count); | |
1066 | |
1067 return di + 1; | |
1068 } | |
1069 | |
1070 /** | |
1071 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4 | |
1072 */ | |
1073 static void encode_rbsp_trailing(PutBitContext *pb){ | |
1074 int length; | |
1075 put_bits(pb, 1, 1); | |
1786 | 1076 length= (-put_bits_count(pb))&7; |
1168 | 1077 if(length) put_bits(pb, length, 0); |
1078 } | |
1079 | |
1080 /** | |
1081 * identifies the exact end of the bitstream | |
1082 * @return the length of the trailing, or 0 if damaged | |
1083 */ | |
1084 static int decode_rbsp_trailing(uint8_t *src){ | |
1085 int v= *src; | |
1086 int r; | |
1087 | |
1170 | 1088 tprintf("rbsp trailing %X\n", v); |
1168 | 1089 |
1090 for(r=1; r<9; r++){ | |
1091 if(v&1) return r; | |
1092 v>>=1; | |
1093 } | |
1094 return 0; | |
1095 } | |
1096 | |
1097 /** | |
1098 * idct tranforms the 16 dc values and dequantize them. | |
1099 * @param qp quantization parameter | |
1100 */ | |
1101 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){ | |
1102 const int qmul= dequant_coeff[qp][0]; | |
1103 #define stride 16 | |
1104 int i; | |
1105 int temp[16]; //FIXME check if this is a good idea | |
1106 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride}; | |
1107 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride}; | |
1108 | |
1109 //memset(block, 64, 2*256); | |
1110 //return; | |
1111 for(i=0; i<4; i++){ | |
1112 const int offset= y_offset[i]; | |
1113 const int z0= block[offset+stride*0] + block[offset+stride*4]; | |
1114 const int z1= block[offset+stride*0] - block[offset+stride*4]; | |
1115 const int z2= block[offset+stride*1] - block[offset+stride*5]; | |
1116 const int z3= block[offset+stride*1] + block[offset+stride*5]; | |
1117 | |
1118 temp[4*i+0]= z0+z3; | |
1119 temp[4*i+1]= z1+z2; | |
1120 temp[4*i+2]= z1-z2; | |
1121 temp[4*i+3]= z0-z3; | |
1122 } | |
1123 | |
1124 for(i=0; i<4; i++){ | |
1125 const int offset= x_offset[i]; | |
1126 const int z0= temp[4*0+i] + temp[4*2+i]; | |
1127 const int z1= temp[4*0+i] - temp[4*2+i]; | |
1128 const int z2= temp[4*1+i] - temp[4*3+i]; | |
1129 const int z3= temp[4*1+i] + temp[4*3+i]; | |
1130 | |
1131 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual | |
1132 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2; | |
1133 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2; | |
1134 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2; | |
1135 } | |
1136 } | |
1137 | |
1138 /** | |
1139 * dct tranforms the 16 dc values. | |
1140 * @param qp quantization parameter ??? FIXME | |
1141 */ | |
1142 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){ | |
1143 // const int qmul= dequant_coeff[qp][0]; | |
1144 int i; | |
1145 int temp[16]; //FIXME check if this is a good idea | |
1146 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride}; | |
1147 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride}; | |
1148 | |
1149 for(i=0; i<4; i++){ | |
1150 const int offset= y_offset[i]; | |
1151 const int z0= block[offset+stride*0] + block[offset+stride*4]; | |
1152 const int z1= block[offset+stride*0] - block[offset+stride*4]; | |
1153 const int z2= block[offset+stride*1] - block[offset+stride*5]; | |
1154 const int z3= block[offset+stride*1] + block[offset+stride*5]; | |
1155 | |
1156 temp[4*i+0]= z0+z3; | |
1157 temp[4*i+1]= z1+z2; | |
1158 temp[4*i+2]= z1-z2; | |
1159 temp[4*i+3]= z0-z3; | |
1160 } | |
1161 | |
1162 for(i=0; i<4; i++){ | |
1163 const int offset= x_offset[i]; | |
1164 const int z0= temp[4*0+i] + temp[4*2+i]; | |
1165 const int z1= temp[4*0+i] - temp[4*2+i]; | |
1166 const int z2= temp[4*1+i] - temp[4*3+i]; | |
1167 const int z3= temp[4*1+i] + temp[4*3+i]; | |
1168 | |
1169 block[stride*0 +offset]= (z0 + z3)>>1; | |
1170 block[stride*2 +offset]= (z1 + z2)>>1; | |
1171 block[stride*8 +offset]= (z1 - z2)>>1; | |
1172 block[stride*10+offset]= (z0 - z3)>>1; | |
1173 } | |
1174 } | |
1175 #undef xStride | |
1176 #undef stride | |
1177 | |
1178 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){ | |
1179 const int qmul= dequant_coeff[qp][0]; | |
1180 const int stride= 16*2; | |
1181 const int xStride= 16; | |
1182 int a,b,c,d,e; | |
1183 | |
1184 a= block[stride*0 + xStride*0]; | |
1185 b= block[stride*0 + xStride*1]; | |
1186 c= block[stride*1 + xStride*0]; | |
1187 d= block[stride*1 + xStride*1]; | |
1188 | |
1189 e= a-b; | |
1190 a= a+b; | |
1191 b= c-d; | |
1192 c= c+d; | |
1193 | |
1194 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1; | |
1195 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1; | |
1196 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1; | |
1197 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1; | |
1198 } | |
1199 | |
1200 static void chroma_dc_dct_c(DCTELEM *block){ | |
1201 const int stride= 16*2; | |
1202 const int xStride= 16; | |
1203 int a,b,c,d,e; | |
1204 | |
1205 a= block[stride*0 + xStride*0]; | |
1206 b= block[stride*0 + xStride*1]; | |
1207 c= block[stride*1 + xStride*0]; | |
1208 d= block[stride*1 + xStride*1]; | |
1209 | |
1210 e= a-b; | |
1211 a= a+b; | |
1212 b= c-d; | |
1213 c= c+d; | |
1214 | |
1215 block[stride*0 + xStride*0]= (a+c); | |
1216 block[stride*0 + xStride*1]= (e+b); | |
1217 block[stride*1 + xStride*0]= (a-c); | |
1218 block[stride*1 + xStride*1]= (e-b); | |
1219 } | |
1220 | |
1221 /** | |
1222 * gets the chroma qp. | |
1223 */ | |
1224 static inline int get_chroma_qp(H264Context *h, int qscale){ | |
1225 | |
1226 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)]; | |
1227 } | |
1228 | |
1229 | |
1230 /** | |
1231 * | |
1232 */ | |
1233 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){ | |
1234 int i; | |
1235 uint8_t *cm = cropTbl + MAX_NEG_CROP; | |
1236 | |
1237 block[0] += 32; | |
1238 #if 1 | |
1239 for(i=0; i<4; i++){ | |
1240 const int z0= block[i + 4*0] + block[i + 4*2]; | |
1241 const int z1= block[i + 4*0] - block[i + 4*2]; | |
1242 const int z2= (block[i + 4*1]>>1) - block[i + 4*3]; | |
1243 const int z3= block[i + 4*1] + (block[i + 4*3]>>1); | |
1244 | |
1245 block[i + 4*0]= z0 + z3; | |
1246 block[i + 4*1]= z1 + z2; | |
1247 block[i + 4*2]= z1 - z2; | |
1248 block[i + 4*3]= z0 - z3; | |
1249 } | |
1250 | |
1251 for(i=0; i<4; i++){ | |
1252 const int z0= block[0 + 4*i] + block[2 + 4*i]; | |
1253 const int z1= block[0 + 4*i] - block[2 + 4*i]; | |
1254 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i]; | |
1255 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1); | |
1256 | |
1257 dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ]; | |
1258 dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ]; | |
1259 dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ]; | |
1260 dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ]; | |
1261 } | |
1262 #else | |
1263 for(i=0; i<4; i++){ | |
1264 const int z0= block[0 + 4*i] + block[2 + 4*i]; | |
1265 const int z1= block[0 + 4*i] - block[2 + 4*i]; | |
1266 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i]; | |
1267 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1); | |
1268 | |
1269 block[0 + 4*i]= z0 + z3; | |
1270 block[1 + 4*i]= z1 + z2; | |
1271 block[2 + 4*i]= z1 - z2; | |
1272 block[3 + 4*i]= z0 - z3; | |
1273 } | |
1274 | |
1275 for(i=0; i<4; i++){ | |
1276 const int z0= block[i + 4*0] + block[i + 4*2]; | |
1277 const int z1= block[i + 4*0] - block[i + 4*2]; | |
1278 const int z2= (block[i + 4*1]>>1) - block[i + 4*3]; | |
1279 const int z3= block[i + 4*1] + (block[i + 4*3]>>1); | |
1280 | |
1281 dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ]; | |
1282 dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ]; | |
1283 dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ]; | |
1284 dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ]; | |
1285 } | |
1286 #endif | |
1287 } | |
1288 | |
1289 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){ | |
1290 int i; | |
1291 //FIXME try int temp instead of block | |
1292 | |
1293 for(i=0; i<4; i++){ | |
1294 const int d0= src1[0 + i*stride] - src2[0 + i*stride]; | |
1295 const int d1= src1[1 + i*stride] - src2[1 + i*stride]; | |
1296 const int d2= src1[2 + i*stride] - src2[2 + i*stride]; | |
1297 const int d3= src1[3 + i*stride] - src2[3 + i*stride]; | |
1298 const int z0= d0 + d3; | |
1299 const int z3= d0 - d3; | |
1300 const int z1= d1 + d2; | |
1301 const int z2= d1 - d2; | |
1302 | |
1303 block[0 + 4*i]= z0 + z1; | |
1304 block[1 + 4*i]= 2*z3 + z2; | |
1305 block[2 + 4*i]= z0 - z1; | |
1306 block[3 + 4*i]= z3 - 2*z2; | |
1307 } | |
1308 | |
1309 for(i=0; i<4; i++){ | |
1310 const int z0= block[0*4 + i] + block[3*4 + i]; | |
1311 const int z3= block[0*4 + i] - block[3*4 + i]; | |
1312 const int z1= block[1*4 + i] + block[2*4 + i]; | |
1313 const int z2= block[1*4 + i] - block[2*4 + i]; | |
1314 | |
1315 block[0*4 + i]= z0 + z1; | |
1316 block[1*4 + i]= 2*z3 + z2; | |
1317 block[2*4 + i]= z0 - z1; | |
1318 block[3*4 + i]= z3 - 2*z2; | |
1319 } | |
1320 } | |
1321 | |
1322 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close | |
1323 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away) | |
1324 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){ | |
1325 int i; | |
1326 const int * const quant_table= quant_coeff[qscale]; | |
1327 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6; | |
1328 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1; | |
1329 const unsigned int threshold2= (threshold1<<1); | |
1330 int last_non_zero; | |
1331 | |
1332 if(seperate_dc){ | |
1333 if(qscale<=18){ | |
1334 //avoid overflows | |
1335 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6; | |
1336 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1; | |
1337 const unsigned int dc_threshold2= (dc_threshold1<<1); | |
1338 | |
1339 int level= block[0]*quant_coeff[qscale+18][0]; | |
1340 if(((unsigned)(level+dc_threshold1))>dc_threshold2){ | |
1341 if(level>0){ | |
1342 level= (dc_bias + level)>>(QUANT_SHIFT-2); | |
1343 block[0]= level; | |
1344 }else{ | |
1345 level= (dc_bias - level)>>(QUANT_SHIFT-2); | |
1346 block[0]= -level; | |
1347 } | |
1348 // last_non_zero = i; | |
1349 }else{ | |
1350 block[0]=0; | |
1351 } | |
1352 }else{ | |
1353 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6; | |
1354 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1; | |
1355 const unsigned int dc_threshold2= (dc_threshold1<<1); | |
1356 | |
1357 int level= block[0]*quant_table[0]; | |
1358 if(((unsigned)(level+dc_threshold1))>dc_threshold2){ | |
1359 if(level>0){ | |
1360 level= (dc_bias + level)>>(QUANT_SHIFT+1); | |
1361 block[0]= level; | |
1362 }else{ | |
1363 level= (dc_bias - level)>>(QUANT_SHIFT+1); | |
1364 block[0]= -level; | |
1365 } | |
1366 // last_non_zero = i; | |
1367 }else{ | |
1368 block[0]=0; | |
1369 } | |
1370 } | |
1371 last_non_zero= 0; | |
1372 i=1; | |
1373 }else{ | |
1374 last_non_zero= -1; | |
1375 i=0; | |
1376 } | |
1377 | |
1378 for(; i<16; i++){ | |
1379 const int j= scantable[i]; | |
1380 int level= block[j]*quant_table[j]; | |
1381 | |
1382 // if( bias+level >= (1<<(QMAT_SHIFT - 3)) | |
1383 // || bias-level >= (1<<(QMAT_SHIFT - 3))){ | |
1384 if(((unsigned)(level+threshold1))>threshold2){ | |
1385 if(level>0){ | |
1386 level= (bias + level)>>QUANT_SHIFT; | |
1387 block[j]= level; | |
1388 }else{ | |
1389 level= (bias - level)>>QUANT_SHIFT; | |
1390 block[j]= -level; | |
1391 } | |
1392 last_non_zero = i; | |
1393 }else{ | |
1394 block[j]=0; | |
1395 } | |
1396 } | |
1397 | |
1398 return last_non_zero; | |
1399 } | |
1400 | |
1401 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){ | |
1402 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
1403 ((uint32_t*)(src+0*stride))[0]= a; | |
1404 ((uint32_t*)(src+1*stride))[0]= a; | |
1405 ((uint32_t*)(src+2*stride))[0]= a; | |
1406 ((uint32_t*)(src+3*stride))[0]= a; | |
1407 } | |
1408 | |
1409 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){ | |
1410 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101; | |
1411 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101; | |
1412 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101; | |
1413 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101; | |
1414 } | |
1415 | |
1416 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1417 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] | |
1418 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3; | |
1419 | |
1420 ((uint32_t*)(src+0*stride))[0]= | |
1421 ((uint32_t*)(src+1*stride))[0]= | |
1422 ((uint32_t*)(src+2*stride))[0]= | |
1423 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1424 } | |
1425 | |
1426 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1427 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2; | |
1428 | |
1429 ((uint32_t*)(src+0*stride))[0]= | |
1430 ((uint32_t*)(src+1*stride))[0]= | |
1431 ((uint32_t*)(src+2*stride))[0]= | |
1432 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1433 } | |
1434 | |
1435 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1436 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2; | |
1437 | |
1438 ((uint32_t*)(src+0*stride))[0]= | |
1439 ((uint32_t*)(src+1*stride))[0]= | |
1440 ((uint32_t*)(src+2*stride))[0]= | |
1441 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1442 } | |
1443 | |
1444 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
1445 ((uint32_t*)(src+0*stride))[0]= | |
1446 ((uint32_t*)(src+1*stride))[0]= | |
1447 ((uint32_t*)(src+2*stride))[0]= | |
1448 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U; | |
1449 } | |
1450 | |
1451 | |
1452 #define LOAD_TOP_RIGHT_EDGE\ | |
1453 const int t4= topright[0];\ | |
1454 const int t5= topright[1];\ | |
1455 const int t6= topright[2];\ | |
1456 const int t7= topright[3];\ | |
1457 | |
1458 #define LOAD_LEFT_EDGE\ | |
1459 const int l0= src[-1+0*stride];\ | |
1460 const int l1= src[-1+1*stride];\ | |
1461 const int l2= src[-1+2*stride];\ | |
1462 const int l3= src[-1+3*stride];\ | |
1463 | |
1464 #define LOAD_TOP_EDGE\ | |
1465 const int t0= src[ 0-1*stride];\ | |
1466 const int t1= src[ 1-1*stride];\ | |
1467 const int t2= src[ 2-1*stride];\ | |
1468 const int t3= src[ 3-1*stride];\ | |
1469 | |
1470 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){ | |
1471 const int lt= src[-1-1*stride]; | |
1472 LOAD_TOP_EDGE | |
1473 LOAD_LEFT_EDGE | |
1474 | |
1475 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; | |
1476 src[0+2*stride]= | |
1477 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; | |
1478 src[0+1*stride]= | |
1479 src[1+2*stride]= | |
1480 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; | |
1481 src[0+0*stride]= | |
1482 src[1+1*stride]= | |
1483 src[2+2*stride]= | |
1484 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
1485 src[1+0*stride]= | |
1486 src[2+1*stride]= | |
1487 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
1488 src[2+0*stride]= | |
1489 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1490 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1282 | 1491 } |
1168 | 1492 |
1493 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){ | |
1494 LOAD_TOP_EDGE | |
1495 LOAD_TOP_RIGHT_EDGE | |
1496 // LOAD_LEFT_EDGE | |
1497 | |
1498 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2; | |
1499 src[1+0*stride]= | |
1500 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2; | |
1501 src[2+0*stride]= | |
1502 src[1+1*stride]= | |
1503 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2; | |
1504 src[3+0*stride]= | |
1505 src[2+1*stride]= | |
1506 src[1+2*stride]= | |
1507 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2; | |
1508 src[3+1*stride]= | |
1509 src[2+2*stride]= | |
1510 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2; | |
1511 src[3+2*stride]= | |
1512 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2; | |
1513 src[3+3*stride]=(t6 + 3*t7 + 2)>>2; | |
1282 | 1514 } |
1168 | 1515 |
1516 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){ | |
1517 const int lt= src[-1-1*stride]; | |
1518 LOAD_TOP_EDGE | |
1519 LOAD_LEFT_EDGE | |
1520 const __attribute__((unused)) int unu= l3; | |
1521 | |
1522 src[0+0*stride]= | |
1523 src[1+2*stride]=(lt + t0 + 1)>>1; | |
1524 src[1+0*stride]= | |
1525 src[2+2*stride]=(t0 + t1 + 1)>>1; | |
1526 src[2+0*stride]= | |
1527 src[3+2*stride]=(t1 + t2 + 1)>>1; | |
1528 src[3+0*stride]=(t2 + t3 + 1)>>1; | |
1529 src[0+1*stride]= | |
1530 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
1531 src[1+1*stride]= | |
1532 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
1533 src[2+1*stride]= | |
1534 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1535 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1536 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2; | |
1537 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1282 | 1538 } |
1168 | 1539 |
1540 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){ | |
1541 LOAD_TOP_EDGE | |
1542 LOAD_TOP_RIGHT_EDGE | |
1543 const __attribute__((unused)) int unu= t7; | |
1544 | |
1545 src[0+0*stride]=(t0 + t1 + 1)>>1; | |
1546 src[1+0*stride]= | |
1547 src[0+2*stride]=(t1 + t2 + 1)>>1; | |
1548 src[2+0*stride]= | |
1549 src[1+2*stride]=(t2 + t3 + 1)>>1; | |
1550 src[3+0*stride]= | |
1551 src[2+2*stride]=(t3 + t4+ 1)>>1; | |
1552 src[3+2*stride]=(t4 + t5+ 1)>>1; | |
1553 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1554 src[1+1*stride]= | |
1555 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1556 src[2+1*stride]= | |
1557 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2; | |
1558 src[3+1*stride]= | |
1559 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2; | |
1560 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2; | |
1282 | 1561 } |
1168 | 1562 |
1563 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){ | |
1564 LOAD_LEFT_EDGE | |
1565 | |
1566 src[0+0*stride]=(l0 + l1 + 1)>>1; | |
1567 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1568 src[2+0*stride]= | |
1569 src[0+1*stride]=(l1 + l2 + 1)>>1; | |
1570 src[3+0*stride]= | |
1571 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2; | |
1572 src[2+1*stride]= | |
1573 src[0+2*stride]=(l2 + l3 + 1)>>1; | |
1574 src[3+1*stride]= | |
1575 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2; | |
1576 src[3+2*stride]= | |
1577 src[1+3*stride]= | |
1578 src[0+3*stride]= | |
1579 src[2+2*stride]= | |
1580 src[2+3*stride]= | |
1581 src[3+3*stride]=l3; | |
1282 | 1582 } |
1168 | 1583 |
1584 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){ | |
1585 const int lt= src[-1-1*stride]; | |
1586 LOAD_TOP_EDGE | |
1587 LOAD_LEFT_EDGE | |
1588 const __attribute__((unused)) int unu= t3; | |
1589 | |
1590 src[0+0*stride]= | |
1591 src[2+1*stride]=(lt + l0 + 1)>>1; | |
1592 src[1+0*stride]= | |
1593 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
1594 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
1595 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
1596 src[0+1*stride]= | |
1597 src[2+2*stride]=(l0 + l1 + 1)>>1; | |
1598 src[1+1*stride]= | |
1599 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2; | |
1600 src[0+2*stride]= | |
1601 src[2+3*stride]=(l1 + l2+ 1)>>1; | |
1602 src[1+2*stride]= | |
1603 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1604 src[0+3*stride]=(l2 + l3 + 1)>>1; | |
1605 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2; | |
1282 | 1606 } |
1168 | 1607 |
1608 static void pred16x16_vertical_c(uint8_t *src, int stride){ | |
1609 int i; | |
1610 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
1611 const uint32_t b= ((uint32_t*)(src-stride))[1]; | |
1612 const uint32_t c= ((uint32_t*)(src-stride))[2]; | |
1613 const uint32_t d= ((uint32_t*)(src-stride))[3]; | |
1614 | |
1615 for(i=0; i<16; i++){ | |
1616 ((uint32_t*)(src+i*stride))[0]= a; | |
1617 ((uint32_t*)(src+i*stride))[1]= b; | |
1618 ((uint32_t*)(src+i*stride))[2]= c; | |
1619 ((uint32_t*)(src+i*stride))[3]= d; | |
1620 } | |
1621 } | |
1622 | |
1623 static void pred16x16_horizontal_c(uint8_t *src, int stride){ | |
1624 int i; | |
1625 | |
1626 for(i=0; i<16; i++){ | |
1627 ((uint32_t*)(src+i*stride))[0]= | |
1628 ((uint32_t*)(src+i*stride))[1]= | |
1629 ((uint32_t*)(src+i*stride))[2]= | |
1630 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101; | |
1631 } | |
1632 } | |
1633 | |
1634 static void pred16x16_dc_c(uint8_t *src, int stride){ | |
1635 int i, dc=0; | |
1636 | |
1637 for(i=0;i<16; i++){ | |
1638 dc+= src[-1+i*stride]; | |
1639 } | |
1640 | |
1641 for(i=0;i<16; i++){ | |
1642 dc+= src[i-stride]; | |
1643 } | |
1644 | |
1645 dc= 0x01010101*((dc + 16)>>5); | |
1646 | |
1647 for(i=0; i<16; i++){ | |
1648 ((uint32_t*)(src+i*stride))[0]= | |
1649 ((uint32_t*)(src+i*stride))[1]= | |
1650 ((uint32_t*)(src+i*stride))[2]= | |
1651 ((uint32_t*)(src+i*stride))[3]= dc; | |
1652 } | |
1653 } | |
1654 | |
1655 static void pred16x16_left_dc_c(uint8_t *src, int stride){ | |
1656 int i, dc=0; | |
1657 | |
1658 for(i=0;i<16; i++){ | |
1659 dc+= src[-1+i*stride]; | |
1660 } | |
1661 | |
1662 dc= 0x01010101*((dc + 8)>>4); | |
1663 | |
1664 for(i=0; i<16; i++){ | |
1665 ((uint32_t*)(src+i*stride))[0]= | |
1666 ((uint32_t*)(src+i*stride))[1]= | |
1667 ((uint32_t*)(src+i*stride))[2]= | |
1668 ((uint32_t*)(src+i*stride))[3]= dc; | |
1669 } | |
1670 } | |
1671 | |
1672 static void pred16x16_top_dc_c(uint8_t *src, int stride){ | |
1673 int i, dc=0; | |
1674 | |
1675 for(i=0;i<16; i++){ | |
1676 dc+= src[i-stride]; | |
1677 } | |
1678 dc= 0x01010101*((dc + 8)>>4); | |
1679 | |
1680 for(i=0; i<16; i++){ | |
1681 ((uint32_t*)(src+i*stride))[0]= | |
1682 ((uint32_t*)(src+i*stride))[1]= | |
1683 ((uint32_t*)(src+i*stride))[2]= | |
1684 ((uint32_t*)(src+i*stride))[3]= dc; | |
1685 } | |
1686 } | |
1687 | |
1688 static void pred16x16_128_dc_c(uint8_t *src, int stride){ | |
1689 int i; | |
1690 | |
1691 for(i=0; i<16; i++){ | |
1692 ((uint32_t*)(src+i*stride))[0]= | |
1693 ((uint32_t*)(src+i*stride))[1]= | |
1694 ((uint32_t*)(src+i*stride))[2]= | |
1695 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U; | |
1696 } | |
1697 } | |
1698 | |
1234 | 1699 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
|
1700 int i, j, k; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1701 int a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1702 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
|
1703 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
|
1704 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
|
1705 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
|
1706 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
|
1707 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
|
1708 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
|
1709 src1 += stride; src2 -= stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1710 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
|
1711 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
|
1712 } |
1234 | 1713 if(svq3){ |
1714 H = ( 5*(H/4) ) / 16; | |
1715 V = ( 5*(V/4) ) / 16; | |
1330
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
1716 |
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
1717 /* required for 100% accuracy */ |
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
1718 i = H; H = V; V = i; |
1234 | 1719 }else{ |
1720 H = ( 5*H+32 ) >> 6; | |
1721 V = ( 5*V+32 ) >> 6; | |
1722 } | |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1723 |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1724 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
|
1725 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
|
1726 int b = a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1727 a += V; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1728 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
|
1729 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
|
1730 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
|
1731 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
|
1732 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
|
1733 b += 4*H; |
1168 | 1734 } |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1735 src += stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1736 } |
1168 | 1737 } |
1738 | |
1234 | 1739 static void pred16x16_plane_c(uint8_t *src, int stride){ |
1740 pred16x16_plane_compat_c(src, stride, 0); | |
1741 } | |
1742 | |
1168 | 1743 static void pred8x8_vertical_c(uint8_t *src, int stride){ |
1744 int i; | |
1745 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
1746 const uint32_t b= ((uint32_t*)(src-stride))[1]; | |
1747 | |
1748 for(i=0; i<8; i++){ | |
1749 ((uint32_t*)(src+i*stride))[0]= a; | |
1750 ((uint32_t*)(src+i*stride))[1]= b; | |
1751 } | |
1752 } | |
1753 | |
1754 static void pred8x8_horizontal_c(uint8_t *src, int stride){ | |
1755 int i; | |
1756 | |
1757 for(i=0; i<8; i++){ | |
1758 ((uint32_t*)(src+i*stride))[0]= | |
1759 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101; | |
1760 } | |
1761 } | |
1762 | |
1763 static void pred8x8_128_dc_c(uint8_t *src, int stride){ | |
1764 int i; | |
1765 | |
1766 for(i=0; i<4; i++){ | |
1767 ((uint32_t*)(src+i*stride))[0]= | |
1768 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U; | |
1769 } | |
1770 for(i=4; i<8; i++){ | |
1771 ((uint32_t*)(src+i*stride))[0]= | |
1772 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U; | |
1773 } | |
1774 } | |
1775 | |
1776 static void pred8x8_left_dc_c(uint8_t *src, int stride){ | |
1777 int i; | |
1778 int dc0, dc2; | |
1779 | |
1780 dc0=dc2=0; | |
1781 for(i=0;i<4; i++){ | |
1782 dc0+= src[-1+i*stride]; | |
1783 dc2+= src[-1+(i+4)*stride]; | |
1784 } | |
1785 dc0= 0x01010101*((dc0 + 2)>>2); | |
1786 dc2= 0x01010101*((dc2 + 2)>>2); | |
1787 | |
1788 for(i=0; i<4; i++){ | |
1789 ((uint32_t*)(src+i*stride))[0]= | |
1790 ((uint32_t*)(src+i*stride))[1]= dc0; | |
1791 } | |
1792 for(i=4; i<8; i++){ | |
1793 ((uint32_t*)(src+i*stride))[0]= | |
1794 ((uint32_t*)(src+i*stride))[1]= dc2; | |
1795 } | |
1796 } | |
1797 | |
1798 static void pred8x8_top_dc_c(uint8_t *src, int stride){ | |
1799 int i; | |
1800 int dc0, dc1; | |
1801 | |
1802 dc0=dc1=0; | |
1803 for(i=0;i<4; i++){ | |
1804 dc0+= src[i-stride]; | |
1805 dc1+= src[4+i-stride]; | |
1806 } | |
1807 dc0= 0x01010101*((dc0 + 2)>>2); | |
1808 dc1= 0x01010101*((dc1 + 2)>>2); | |
1809 | |
1810 for(i=0; i<4; i++){ | |
1811 ((uint32_t*)(src+i*stride))[0]= dc0; | |
1812 ((uint32_t*)(src+i*stride))[1]= dc1; | |
1813 } | |
1814 for(i=4; i<8; i++){ | |
1815 ((uint32_t*)(src+i*stride))[0]= dc0; | |
1816 ((uint32_t*)(src+i*stride))[1]= dc1; | |
1817 } | |
1818 } | |
1819 | |
1820 | |
1821 static void pred8x8_dc_c(uint8_t *src, int stride){ | |
1822 int i; | |
1823 int dc0, dc1, dc2, dc3; | |
1824 | |
1825 dc0=dc1=dc2=0; | |
1826 for(i=0;i<4; i++){ | |
1827 dc0+= src[-1+i*stride] + src[i-stride]; | |
1828 dc1+= src[4+i-stride]; | |
1829 dc2+= src[-1+(i+4)*stride]; | |
1830 } | |
1831 dc3= 0x01010101*((dc1 + dc2 + 4)>>3); | |
1832 dc0= 0x01010101*((dc0 + 4)>>3); | |
1833 dc1= 0x01010101*((dc1 + 2)>>2); | |
1834 dc2= 0x01010101*((dc2 + 2)>>2); | |
1835 | |
1836 for(i=0; i<4; i++){ | |
1837 ((uint32_t*)(src+i*stride))[0]= dc0; | |
1838 ((uint32_t*)(src+i*stride))[1]= dc1; | |
1839 } | |
1840 for(i=4; i<8; i++){ | |
1841 ((uint32_t*)(src+i*stride))[0]= dc2; | |
1842 ((uint32_t*)(src+i*stride))[1]= dc3; | |
1843 } | |
1844 } | |
1845 | |
1846 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
|
1847 int j, k; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1848 int a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1849 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
|
1850 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
|
1851 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
|
1852 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
|
1853 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
|
1854 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
|
1855 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
|
1856 src1 += stride; src2 -= stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1857 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
|
1858 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
|
1859 } |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1860 H = ( 17*H+16 ) >> 5; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1861 V = ( 17*V+16 ) >> 5; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1862 |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1863 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
|
1864 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
|
1865 int b = a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1866 a += V; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1867 src[0] = cm[ (b ) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1868 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
|
1869 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
|
1870 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
|
1871 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
|
1872 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
|
1873 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
|
1874 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
|
1875 src += stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
1876 } |
1168 | 1877 } |
1878 | |
1879 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list, | |
1880 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
1881 int src_x_offset, int src_y_offset, | |
1882 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){ | |
1883 MpegEncContext * const s = &h->s; | |
1884 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8; | |
1885 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8; | |
1886 const int luma_xy= (mx&3) + ((my&3)<<2); | |
1887 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize; | |
1888 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize; | |
1889 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize; | |
1890 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it | |
1891 int extra_height= extra_width; | |
1892 int emu=0; | |
1893 const int full_mx= mx>>2; | |
1894 const int full_my= my>>2; | |
1895 | |
1896 assert(pic->data[0]); | |
1897 | |
1898 if(mx&7) extra_width -= 3; | |
1899 if(my&7) extra_height -= 3; | |
1900 | |
1901 if( full_mx < 0-extra_width | |
1902 || full_my < 0-extra_height | |
1903 || full_mx + 16/*FIXME*/ > s->width + extra_width | |
1904 || full_my + 16/*FIXME*/ > s->height + extra_height){ | |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1282
diff
changeset
|
1905 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 | 1906 src_y= s->edge_emu_buffer + 2 + 2*s->linesize; |
1907 emu=1; | |
1908 } | |
1909 | |
1910 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps? | |
1911 if(!square){ | |
1912 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize); | |
1913 } | |
1914 | |
1915 if(s->flags&CODEC_FLAG_GRAY) return; | |
1916 | |
1917 if(emu){ | |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1282
diff
changeset
|
1918 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 | 1919 src_cb= s->edge_emu_buffer; |
1920 } | |
1921 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7); | |
1922 | |
1923 if(emu){ | |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1282
diff
changeset
|
1924 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 | 1925 src_cr= s->edge_emu_buffer; |
1926 } | |
1927 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7); | |
1928 } | |
1929 | |
1930 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta, | |
1931 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
1932 int x_offset, int y_offset, | |
1933 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put, | |
1934 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg, | |
1935 int list0, int list1){ | |
1936 MpegEncContext * const s = &h->s; | |
1937 qpel_mc_func *qpix_op= qpix_put; | |
1938 h264_chroma_mc_func chroma_op= chroma_put; | |
1939 | |
1940 dest_y += 2*x_offset + 2*y_offset*s-> linesize; | |
1941 dest_cb += x_offset + y_offset*s->uvlinesize; | |
1942 dest_cr += x_offset + y_offset*s->uvlinesize; | |
1943 x_offset += 8*s->mb_x; | |
1944 y_offset += 8*s->mb_y; | |
1945 | |
1946 if(list0){ | |
1169 | 1947 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ]; |
1168 | 1948 mc_dir_part(h, ref, n, square, chroma_height, delta, 0, |
1949 dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
1950 qpix_op, chroma_op); | |
1951 | |
1952 qpix_op= qpix_avg; | |
1953 chroma_op= chroma_avg; | |
1954 } | |
1955 | |
1956 if(list1){ | |
1169 | 1957 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ]; |
1168 | 1958 mc_dir_part(h, ref, n, square, chroma_height, delta, 1, |
1959 dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
1960 qpix_op, chroma_op); | |
1961 } | |
1962 } | |
1963 | |
1964 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
1965 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put), | |
1966 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){ | |
1967 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
|
1968 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 1969 const int mb_type= s->current_picture.mb_type[mb_xy]; |
1970 | |
1971 assert(IS_INTER(mb_type)); | |
1972 | |
1973 if(IS_16X16(mb_type)){ | |
1974 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0, | |
1975 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0], | |
1976 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); | |
1977 }else if(IS_16X8(mb_type)){ | |
1978 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0, | |
1979 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0], | |
1980 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); | |
1981 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4, | |
1982 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0], | |
1983 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1)); | |
1984 }else if(IS_8X16(mb_type)){ | |
1985 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0, | |
1986 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], | |
1987 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); | |
1988 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0, | |
1989 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], | |
1990 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1)); | |
1991 }else{ | |
1992 int i; | |
1993 | |
1994 assert(IS_8X8(mb_type)); | |
1995 | |
1996 for(i=0; i<4; i++){ | |
1997 const int sub_mb_type= h->sub_mb_type[i]; | |
1998 const int n= 4*i; | |
1999 int x_offset= (i&1)<<2; | |
2000 int y_offset= (i&2)<<1; | |
2001 | |
2002 if(IS_SUB_8X8(sub_mb_type)){ | |
2003 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2004 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], | |
2005 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2006 }else if(IS_SUB_8X4(sub_mb_type)){ | |
2007 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2008 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1], | |
2009 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2010 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2, | |
2011 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1], | |
2012 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2013 }else if(IS_SUB_4X8(sub_mb_type)){ | |
2014 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2015 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], | |
2016 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2017 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset, | |
2018 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], | |
2019 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2020 }else{ | |
2021 int j; | |
2022 assert(IS_SUB_4X4(sub_mb_type)); | |
2023 for(j=0; j<4; j++){ | |
2024 int sub_x_offset= x_offset + 2*(j&1); | |
2025 int sub_y_offset= y_offset + (j&2); | |
2026 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset, | |
2027 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], | |
2028 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); | |
2029 } | |
2030 } | |
2031 } | |
2032 } | |
2033 } | |
2034 | |
2035 static void decode_init_vlc(H264Context *h){ | |
2036 static int done = 0; | |
2037 | |
2038 if (!done) { | |
2039 int i; | |
2040 done = 1; | |
2041 | |
2042 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, | |
2043 &chroma_dc_coeff_token_len [0], 1, 1, | |
2044 &chroma_dc_coeff_token_bits[0], 1, 1); | |
2045 | |
2046 for(i=0; i<4; i++){ | |
2047 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, | |
2048 &coeff_token_len [i][0], 1, 1, | |
2049 &coeff_token_bits[i][0], 1, 1); | |
2050 } | |
2051 | |
2052 for(i=0; i<3; i++){ | |
2053 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4, | |
2054 &chroma_dc_total_zeros_len [i][0], 1, 1, | |
2055 &chroma_dc_total_zeros_bits[i][0], 1, 1); | |
2056 } | |
2057 for(i=0; i<15; i++){ | |
2058 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, | |
2059 &total_zeros_len [i][0], 1, 1, | |
2060 &total_zeros_bits[i][0], 1, 1); | |
2061 } | |
2062 | |
2063 for(i=0; i<6; i++){ | |
2064 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, | |
2065 &run_len [i][0], 1, 1, | |
2066 &run_bits[i][0], 1, 1); | |
2067 } | |
2068 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, | |
2069 &run_len [6][0], 1, 1, | |
2070 &run_bits[6][0], 1, 1); | |
2071 } | |
2072 } | |
2073 | |
2074 /** | |
2075 * Sets the intra prediction function pointers. | |
2076 */ | |
2077 static void init_pred_ptrs(H264Context *h){ | |
2078 // MpegEncContext * const s = &h->s; | |
2079 | |
2080 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c; | |
2081 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c; | |
2082 h->pred4x4[DC_PRED ]= pred4x4_dc_c; | |
2083 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c; | |
2084 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c; | |
2085 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c; | |
2086 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c; | |
2087 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c; | |
2088 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c; | |
2089 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c; | |
2090 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c; | |
2091 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c; | |
2092 | |
2093 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c; | |
2094 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c; | |
2095 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c; | |
2096 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c; | |
2097 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c; | |
2098 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c; | |
2099 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c; | |
2100 | |
2101 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c; | |
2102 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c; | |
2103 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c; | |
2104 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c; | |
2105 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c; | |
2106 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c; | |
2107 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c; | |
2108 } | |
2109 | |
2110 static void free_tables(H264Context *h){ | |
2111 av_freep(&h->intra4x4_pred_mode); | |
2112 av_freep(&h->non_zero_count); | |
2113 av_freep(&h->slice_table_base); | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2114 av_freep(&h->top_border); |
1168 | 2115 h->slice_table= NULL; |
2116 | |
2117 av_freep(&h->mb2b_xy); | |
2118 av_freep(&h->mb2b8_xy); | |
2119 } | |
2120 | |
2121 /** | |
2122 * allocates tables. | |
2123 * needs widzh/height | |
2124 */ | |
2125 static int alloc_tables(H264Context *h){ | |
2126 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
|
2127 const int big_mb_num= s->mb_stride * (s->mb_height+1); |
1168 | 2128 int x,y; |
2129 | |
2130 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t)) | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2131 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t)) |
1168 | 2132 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t)) |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2133 CHECKED_ALLOCZ(h->top_border , s->mb_width * (16+8+8) * sizeof(uint8_t)) |
1168 | 2134 |
2135 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
|
2136 h->slice_table= h->slice_table_base + s->mb_stride + 1; |
1168 | 2137 |
2138 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t)); | |
2139 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t)); | |
2140 for(y=0; y<s->mb_height; y++){ | |
2141 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
|
2142 const int mb_xy= x + y*s->mb_stride; |
1168 | 2143 const int b_xy = 4*x + 4*y*h->b_stride; |
2144 const int b8_xy= 2*x + 2*y*h->b8_stride; | |
2145 | |
2146 h->mb2b_xy [mb_xy]= b_xy; | |
2147 h->mb2b8_xy[mb_xy]= b8_xy; | |
2148 } | |
2149 } | |
2150 | |
2151 return 0; | |
2152 fail: | |
2153 free_tables(h); | |
2154 return -1; | |
2155 } | |
2156 | |
2157 static void common_init(H264Context *h){ | |
2158 MpegEncContext * const s = &h->s; | |
2159 | |
2160 s->width = s->avctx->width; | |
2161 s->height = s->avctx->height; | |
2162 s->codec_id= s->avctx->codec->id; | |
2163 | |
2164 init_pred_ptrs(h); | |
2165 | |
1698 | 2166 s->unrestricted_mv=1; |
1168 | 2167 s->decode=1; //FIXME |
2168 } | |
2169 | |
2170 static int decode_init(AVCodecContext *avctx){ | |
2171 H264Context *h= avctx->priv_data; | |
2172 MpegEncContext * const s = &h->s; | |
2173 | |
1892 | 2174 MPV_decode_defaults(s); |
2175 | |
1168 | 2176 s->avctx = avctx; |
2177 common_init(h); | |
2178 | |
2179 s->out_format = FMT_H264; | |
2180 s->workaround_bugs= avctx->workaround_bugs; | |
2181 | |
2182 // set defaults | |
2183 // s->decode_mb= ff_h263_decode_mb; | |
2184 s->low_delay= 1; | |
2185 avctx->pix_fmt= PIX_FMT_YUV420P; | |
2186 | |
2187 decode_init_vlc(h); | |
2188 | |
2189 return 0; | |
2190 } | |
2191 | |
2192 static void frame_start(H264Context *h){ | |
2193 MpegEncContext * const s = &h->s; | |
2194 int i; | |
2195 | |
2196 MPV_frame_start(s, s->avctx); | |
2197 ff_er_frame_start(s); | |
2198 h->mmco_index=0; | |
2199 | |
2200 assert(s->linesize && s->uvlinesize); | |
2201 | |
2202 for(i=0; i<16; i++){ | |
2203 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3); | |
2204 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3); | |
2205 } | |
2206 for(i=0; i<4; i++){ | |
2207 h->block_offset[16+i]= | |
2208 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3); | |
2209 } | |
2210 | |
2211 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1; | |
2212 } | |
2213 | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2214 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2215 MpegEncContext * const s = &h->s; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2216 int i; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2217 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2218 src_y -= linesize; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2219 src_cb -= uvlinesize; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2220 src_cr -= uvlinesize; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2221 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2222 h->left_border[0]= h->top_border[s->mb_x][15]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2223 for(i=1; i<17; i++){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2224 h->left_border[i]= src_y[15+i* linesize]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2225 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2226 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2227 *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2228 *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2229 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2230 if(!(s->flags&CODEC_FLAG_GRAY)){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2231 h->left_border[17 ]= h->top_border[s->mb_x][16+7]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2232 h->left_border[17+9]= h->top_border[s->mb_x][24+7]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2233 for(i=1; i<9; i++){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2234 h->left_border[i+17 ]= src_cb[7+i*uvlinesize]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2235 h->left_border[i+17+9]= src_cr[7+i*uvlinesize]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2236 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2237 *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2238 *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2239 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2240 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2241 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2242 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2243 MpegEncContext * const s = &h->s; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2244 int temp8, i; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2245 uint64_t temp64; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2246 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2247 src_y -= linesize + 1; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2248 src_cb -= uvlinesize + 1; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2249 src_cr -= uvlinesize + 1; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2250 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2251 #define XCHG(a,b,t,xchg)\ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2252 t= a;\ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2253 if(xchg)\ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2254 a= b;\ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2255 b= t; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2256 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2257 for(i=0; i<17; i++){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2258 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2259 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2260 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2261 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2262 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2263 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2264 if(!(s->flags&CODEC_FLAG_GRAY)){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2265 for(i=0; i<9; i++){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2266 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2267 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2268 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2269 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2270 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2271 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2272 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2273 |
1168 | 2274 static void hl_decode_mb(H264Context *h){ |
2275 MpegEncContext * const s = &h->s; | |
2276 const int mb_x= s->mb_x; | |
2277 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
|
2278 const int mb_xy= mb_x + mb_y*s->mb_stride; |
1168 | 2279 const int mb_type= s->current_picture.mb_type[mb_xy]; |
2280 uint8_t *dest_y, *dest_cb, *dest_cr; | |
2281 int linesize, uvlinesize /*dct_offset*/; | |
2282 int i; | |
2283 | |
2284 if(!s->decode) | |
2285 return; | |
2286 | |
2287 if(s->mb_skiped){ | |
2288 } | |
2289 | |
2290 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16; | |
2291 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; | |
2292 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; | |
2293 | |
2294 if (h->mb_field_decoding_flag) { | |
2295 linesize = s->linesize * 2; | |
2296 uvlinesize = s->uvlinesize * 2; | |
2297 if(mb_y&1){ //FIXME move out of this func? | |
2298 dest_y -= s->linesize*15; | |
2299 dest_cb-= s->linesize*7; | |
2300 dest_cr-= s->linesize*7; | |
2301 } | |
2302 } else { | |
2303 linesize = s->linesize; | |
2304 uvlinesize = s->uvlinesize; | |
2305 // dct_offset = s->linesize * 16; | |
2306 } | |
2307 | |
2308 if(IS_INTRA(mb_type)){ | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2309 if(h->deblocking_filter) |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2310 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2311 |
1168 | 2312 if(!(s->flags&CODEC_FLAG_GRAY)){ |
2313 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize); | |
2314 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize); | |
2315 } | |
2316 | |
2317 if(IS_INTRA4x4(mb_type)){ | |
2318 if(!s->encoding){ | |
2319 for(i=0; i<16; i++){ | |
2320 uint8_t * const ptr= dest_y + h->block_offset[i]; | |
2321 uint8_t *topright= ptr + 4 - linesize; | |
2322 const int topright_avail= (h->topright_samples_available<<i)&0x8000; | |
2323 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ]; | |
2324 int tr; | |
2325 | |
2326 if(!topright_avail){ | |
2327 tr= ptr[3 - linesize]*0x01010101; | |
2328 topright= (uint8_t*) &tr; | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2329 }else if(i==5 && h->deblocking_filter){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2330 tr= *(uint32_t*)h->top_border[mb_x+1]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2331 topright= (uint8_t*) &tr; |
1168 | 2332 } |
2333 | |
2334 h->pred4x4[ dir ](ptr, topright, linesize); | |
1234 | 2335 if(h->non_zero_count_cache[ scan8[i] ]){ |
2336 if(s->codec_id == CODEC_ID_H264) | |
2337 h264_add_idct_c(ptr, h->mb + i*16, linesize); | |
2338 else | |
2339 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0); | |
2340 } | |
1168 | 2341 } |
2342 } | |
2343 }else{ | |
2344 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize); | |
1234 | 2345 if(s->codec_id == CODEC_ID_H264) |
2346 h264_luma_dc_dequant_idct_c(h->mb, s->qscale); | |
2347 else | |
2348 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale); | |
1168 | 2349 } |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2350 if(h->deblocking_filter) |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2351 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0); |
1234 | 2352 }else if(s->codec_id == CODEC_ID_H264){ |
1168 | 2353 hl_motion(h, dest_y, dest_cb, dest_cr, |
2354 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, | |
2355 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab); | |
2356 } | |
2357 | |
2358 | |
2359 if(!IS_INTRA4x4(mb_type)){ | |
1250 | 2360 if(s->codec_id == CODEC_ID_H264){ |
2361 for(i=0; i<16; i++){ | |
2362 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below | |
2363 uint8_t * const ptr= dest_y + h->block_offset[i]; | |
1234 | 2364 h264_add_idct_c(ptr, h->mb + i*16, linesize); |
1250 | 2365 } |
2366 } | |
2367 }else{ | |
2368 for(i=0; i<16; i++){ | |
2369 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below | |
2370 uint8_t * const ptr= dest_y + h->block_offset[i]; | |
1234 | 2371 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0); |
1250 | 2372 } |
1168 | 2373 } |
2374 } | |
2375 } | |
2376 | |
2377 if(!(s->flags&CODEC_FLAG_GRAY)){ | |
2378 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp); | |
2379 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp); | |
1250 | 2380 if(s->codec_id == CODEC_ID_H264){ |
2381 for(i=16; i<16+4; i++){ | |
2382 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ | |
2383 uint8_t * const ptr= dest_cb + h->block_offset[i]; | |
1234 | 2384 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize); |
1250 | 2385 } |
2386 } | |
2387 for(i=20; i<20+4; i++){ | |
2388 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ | |
2389 uint8_t * const ptr= dest_cr + h->block_offset[i]; | |
2390 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize); | |
2391 } | |
1168 | 2392 } |
1250 | 2393 }else{ |
2394 for(i=16; i<16+4; i++){ | |
2395 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ | |
2396 uint8_t * const ptr= dest_cb + h->block_offset[i]; | |
1234 | 2397 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2); |
1250 | 2398 } |
2399 } | |
2400 for(i=20; i<20+4; i++){ | |
2401 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ | |
2402 uint8_t * const ptr= dest_cr + h->block_offset[i]; | |
2403 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2); | |
2404 } | |
1168 | 2405 } |
2406 } | |
2407 } | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2408 if(h->deblocking_filter) { |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2409 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2410 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
2411 } |
1168 | 2412 } |
2413 | |
2414 static void decode_mb_cabac(H264Context *h){ | |
2415 // MpegEncContext * const s = &h->s; | |
2416 } | |
2417 | |
2418 /** | |
2419 * fills the default_ref_list. | |
2420 */ | |
2421 static int fill_default_ref_list(H264Context *h){ | |
2422 MpegEncContext * const s = &h->s; | |
2423 int i; | |
2424 Picture sorted_short_ref[16]; | |
2425 | |
2426 if(h->slice_type==B_TYPE){ | |
2427 int out_i; | |
2428 int limit= -1; | |
2429 | |
2430 for(out_i=0; out_i<h->short_ref_count; out_i++){ | |
2431 int best_i=-1; | |
2432 int best_poc=-1; | |
2433 | |
2434 for(i=0; i<h->short_ref_count; i++){ | |
2435 const int poc= h->short_ref[i]->poc; | |
2436 if(poc > limit && poc < best_poc){ | |
2437 best_poc= poc; | |
2438 best_i= i; | |
2439 } | |
2440 } | |
2441 | |
2442 assert(best_i != -1); | |
2443 | |
2444 limit= best_poc; | |
2445 sorted_short_ref[out_i]= *h->short_ref[best_i]; | |
2446 } | |
2447 } | |
2448 | |
2449 if(s->picture_structure == PICT_FRAME){ | |
2450 if(h->slice_type==B_TYPE){ | |
2451 const int current_poc= s->current_picture_ptr->poc; | |
2452 int list; | |
2453 | |
2454 for(list=0; list<2; list++){ | |
2455 int index=0; | |
2456 | |
2457 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){ | |
2458 const int i2= list ? h->short_ref_count - i - 1 : i; | |
2459 const int poc= sorted_short_ref[i2].poc; | |
2460 | |
2461 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit | |
2462 | |
2463 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){ | |
2464 h->default_ref_list[list][index ]= sorted_short_ref[i2]; | |
2465 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num; | |
2466 } | |
2467 } | |
2468 | |
2469 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){ | |
2470 if(h->long_ref[i]->reference != 3) continue; | |
2471 | |
2472 h->default_ref_list[ list ][index ]= *h->long_ref[i]; | |
2473 h->default_ref_list[ list ][index++].pic_id= i;; | |
2474 } | |
2475 | |
2476 if(h->long_ref_count > 1 && h->short_ref_count==0){ | |
2477 Picture temp= h->default_ref_list[1][0]; | |
2478 h->default_ref_list[1][0] = h->default_ref_list[1][1]; | |
2479 h->default_ref_list[1][0] = temp; | |
2480 } | |
2481 | |
2482 if(index < h->ref_count[ list ]) | |
2483 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index)); | |
2484 } | |
2485 }else{ | |
2486 int index=0; | |
2487 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){ | |
2488 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit | |
2489 h->default_ref_list[0][index ]= *h->short_ref[i]; | |
2490 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num; | |
2491 } | |
2492 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){ | |
2493 if(h->long_ref[i]->reference != 3) continue; | |
2494 h->default_ref_list[0][index ]= *h->long_ref[i]; | |
2495 h->default_ref_list[0][index++].pic_id= i;; | |
2496 } | |
2497 if(index < h->ref_count[0]) | |
2498 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index)); | |
2499 } | |
2500 }else{ //FIELD | |
2501 if(h->slice_type==B_TYPE){ | |
2502 }else{ | |
2503 //FIXME second field balh | |
2504 } | |
2505 } | |
2506 return 0; | |
2507 } | |
2508 | |
2509 static int decode_ref_pic_list_reordering(H264Context *h){ | |
2510 MpegEncContext * const s = &h->s; | |
2511 int list; | |
2512 | |
2513 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func | |
2514 | |
2515 for(list=0; list<2; list++){ | |
2516 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]); | |
2517 | |
2518 if(get_bits1(&s->gb)){ | |
2519 int pred= h->curr_pic_num; | |
2520 int index; | |
2521 | |
2522 for(index=0; ; index++){ | |
2523 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb); | |
2524 int pic_id; | |
2525 int i; | |
2526 | |
2527 | |
2528 if(index >= h->ref_count[list]){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2529 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n"); |
1168 | 2530 return -1; |
2531 } | |
2532 | |
2533 if(reordering_of_pic_nums_idc<3){ | |
2534 if(reordering_of_pic_nums_idc<2){ | |
2535 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1; | |
2536 | |
2537 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
|
2538 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n"); |
1168 | 2539 return -1; |
2540 } | |
2541 | |
2542 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num; | |
2543 else pred+= abs_diff_pic_num; | |
2544 pred &= h->max_pic_num - 1; | |
2545 | |
2546 for(i= h->ref_count[list]-1; i>=index; i--){ | |
2547 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0) | |
2548 break; | |
2549 } | |
2550 }else{ | |
2551 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx | |
2552 | |
2553 for(i= h->ref_count[list]-1; i>=index; i--){ | |
2554 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1) | |
2555 break; | |
2556 } | |
2557 } | |
2558 | |
2559 if(i < index){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2560 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n"); |
1168 | 2561 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME |
2562 }else if(i > index){ | |
2563 Picture tmp= h->ref_list[list][i]; | |
2564 for(; i>index; i--){ | |
2565 h->ref_list[list][i]= h->ref_list[list][i-1]; | |
2566 } | |
2567 h->ref_list[list][index]= tmp; | |
2568 } | |
2569 }else if(reordering_of_pic_nums_idc==3) | |
2570 break; | |
2571 else{ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2572 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n"); |
1168 | 2573 return -1; |
2574 } | |
2575 } | |
2576 } | |
2577 | |
2578 if(h->slice_type!=B_TYPE) break; | |
2579 } | |
2580 return 0; | |
2581 } | |
2582 | |
2583 static int pred_weight_table(H264Context *h){ | |
2584 MpegEncContext * const s = &h->s; | |
2585 int list, i; | |
2586 | |
2587 h->luma_log2_weight_denom= get_ue_golomb(&s->gb); | |
2588 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb); | |
2589 | |
2590 for(list=0; list<2; list++){ | |
2591 for(i=0; i<h->ref_count[list]; i++){ | |
2592 int luma_weight_flag, chroma_weight_flag; | |
2593 | |
2594 luma_weight_flag= get_bits1(&s->gb); | |
2595 if(luma_weight_flag){ | |
2596 h->luma_weight[list][i]= get_se_golomb(&s->gb); | |
2597 h->luma_offset[list][i]= get_se_golomb(&s->gb); | |
2598 } | |
2599 | |
2600 chroma_weight_flag= get_bits1(&s->gb); | |
2601 if(chroma_weight_flag){ | |
2602 int j; | |
2603 for(j=0; j<2; j++){ | |
2604 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb); | |
2605 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb); | |
2606 } | |
2607 } | |
2608 } | |
2609 if(h->slice_type != B_TYPE) break; | |
2610 } | |
2611 return 0; | |
2612 } | |
2613 | |
2614 /** | |
2615 * instantaneos decoder refresh. | |
2616 */ | |
2617 static void idr(H264Context *h){ | |
2618 int i; | |
2619 | |
2620 for(i=0; i<h->long_ref_count; i++){ | |
2621 h->long_ref[i]->reference=0; | |
2622 h->long_ref[i]= NULL; | |
2623 } | |
2624 h->long_ref_count=0; | |
2625 | |
2626 for(i=0; i<h->short_ref_count; i++){ | |
2627 h->short_ref[i]->reference=0; | |
2628 h->short_ref[i]= NULL; | |
2629 } | |
2630 h->short_ref_count=0; | |
2631 } | |
2632 | |
2633 /** | |
2634 * | |
2635 * @return the removed picture or NULL if an error occures | |
2636 */ | |
2637 static Picture * remove_short(H264Context *h, int frame_num){ | |
1169 | 2638 MpegEncContext * const s = &h->s; |
1168 | 2639 int i; |
2640 | |
1169 | 2641 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
|
2642 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count); |
1169 | 2643 |
1168 | 2644 for(i=0; i<h->short_ref_count; i++){ |
2645 Picture *pic= h->short_ref[i]; | |
1169 | 2646 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
|
2647 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic); |
1168 | 2648 if(pic->frame_num == frame_num){ |
2649 h->short_ref[i]= NULL; | |
2650 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*)); | |
2651 h->short_ref_count--; | |
2652 return pic; | |
2653 } | |
2654 } | |
2655 return NULL; | |
2656 } | |
2657 | |
2658 /** | |
2659 * | |
2660 * @return the removed picture or NULL if an error occures | |
2661 */ | |
2662 static Picture * remove_long(H264Context *h, int i){ | |
2663 Picture *pic; | |
2664 | |
2665 if(i >= h->long_ref_count) return NULL; | |
2666 pic= h->long_ref[i]; | |
2667 if(pic==NULL) return NULL; | |
2668 | |
2669 h->long_ref[i]= NULL; | |
2670 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*)); | |
2671 h->long_ref_count--; | |
2672 | |
2673 return pic; | |
2674 } | |
2675 | |
2676 /** | |
2677 * Executes the reference picture marking (memory management control operations). | |
2678 */ | |
2679 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){ | |
2680 MpegEncContext * const s = &h->s; | |
2681 int i; | |
2682 int current_is_long=0; | |
2683 Picture *pic; | |
2684 | |
2685 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
|
2686 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n"); |
1168 | 2687 |
2688 for(i=0; i<mmco_count; i++){ | |
2689 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
|
2690 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 | 2691 |
2692 switch(mmco[i].opcode){ | |
2693 case MMCO_SHORT2UNUSED: | |
2694 pic= remove_short(h, mmco[i].short_frame_num); | |
2695 if(pic==NULL) return -1; | |
2696 pic->reference= 0; | |
2697 break; | |
2698 case MMCO_SHORT2LONG: | |
2699 pic= remove_long(h, mmco[i].long_index); | |
2700 if(pic) pic->reference=0; | |
2701 | |
2702 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num); | |
2703 h->long_ref[ mmco[i].long_index ]->long_ref=1; | |
2704 break; | |
2705 case MMCO_LONG2UNUSED: | |
2706 pic= remove_long(h, mmco[i].long_index); | |
2707 if(pic==NULL) return -1; | |
2708 pic->reference= 0; | |
2709 break; | |
2710 case MMCO_LONG: | |
2711 pic= remove_long(h, mmco[i].long_index); | |
2712 if(pic) pic->reference=0; | |
2713 | |
2714 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr; | |
2715 h->long_ref[ mmco[i].long_index ]->long_ref=1; | |
2716 h->long_ref_count++; | |
2717 | |
2718 current_is_long=1; | |
2719 break; | |
2720 case MMCO_SET_MAX_LONG: | |
2721 assert(mmco[i].long_index <= 16); | |
2722 while(mmco[i].long_index < h->long_ref_count){ | |
2723 pic= remove_long(h, mmco[i].long_index); | |
2724 pic->reference=0; | |
2725 } | |
2726 while(mmco[i].long_index > h->long_ref_count){ | |
2727 h->long_ref[ h->long_ref_count++ ]= NULL; | |
2728 } | |
2729 break; | |
2730 case MMCO_RESET: | |
2731 while(h->short_ref_count){ | |
2732 pic= remove_short(h, h->short_ref[0]->frame_num); | |
2733 pic->reference=0; | |
2734 } | |
2735 while(h->long_ref_count){ | |
2736 pic= remove_long(h, h->long_ref_count-1); | |
2737 pic->reference=0; | |
2738 } | |
2739 break; | |
2740 default: assert(0); | |
2741 } | |
2742 } | |
2743 | |
2744 if(!current_is_long){ | |
2745 pic= remove_short(h, s->current_picture_ptr->frame_num); | |
2746 if(pic){ | |
2747 pic->reference=0; | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2748 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n"); |
1168 | 2749 } |
2750 | |
2751 if(h->short_ref_count) | |
1169 | 2752 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*)); |
2753 | |
2754 h->short_ref[0]= s->current_picture_ptr; | |
1168 | 2755 h->short_ref[0]->long_ref=0; |
2756 h->short_ref_count++; | |
2757 } | |
2758 | |
2759 return 0; | |
2760 } | |
2761 | |
2762 static int decode_ref_pic_marking(H264Context *h){ | |
2763 MpegEncContext * const s = &h->s; | |
2764 int i; | |
2765 | |
2766 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields | |
2767 s->broken_link= get_bits1(&s->gb) -1; | |
2768 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx | |
2769 if(h->mmco[0].long_index == -1) | |
2770 h->mmco_index= 0; | |
2771 else{ | |
2772 h->mmco[0].opcode= MMCO_LONG; | |
2773 h->mmco_index= 1; | |
2774 } | |
2775 }else{ | |
2776 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag | |
2777 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) { | |
2778 MMCOOpcode opcode= get_ue_golomb(&s->gb);; | |
2779 | |
2780 h->mmco[i].opcode= opcode; | |
2781 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){ | |
2782 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 | |
2783 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){ | |
2784 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco); | |
2785 return -1; | |
2786 }*/ | |
2787 } | |
2788 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){ | |
2789 h->mmco[i].long_index= get_ue_golomb(&s->gb); | |
2790 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
|
2791 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode); |
1168 | 2792 return -1; |
2793 } | |
2794 } | |
2795 | |
2796 if(opcode > MMCO_LONG){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2797 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode); |
1168 | 2798 return -1; |
2799 } | |
2800 } | |
2801 h->mmco_index= i; | |
2802 }else{ | |
2803 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count); | |
2804 | |
2805 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields | |
2806 h->mmco[0].opcode= MMCO_SHORT2UNUSED; | |
2807 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num; | |
2808 h->mmco_index= 1; | |
2809 }else | |
2810 h->mmco_index= 0; | |
2811 } | |
2812 } | |
2813 | |
2814 return 0; | |
2815 } | |
2816 | |
2817 static int init_poc(H264Context *h){ | |
2818 MpegEncContext * const s = &h->s; | |
2819 const int max_frame_num= 1<<h->sps.log2_max_frame_num; | |
2820 int field_poc[2]; | |
2821 | |
2822 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
2823 h->frame_num_offset= 0; | |
2824 }else{ | |
2825 if(h->frame_num < h->prev_frame_num) | |
2826 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num; | |
2827 else | |
2828 h->frame_num_offset= h->prev_frame_num_offset; | |
2829 } | |
2830 | |
2831 if(h->sps.poc_type==0){ | |
2832 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb; | |
2833 | |
2834 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2) | |
2835 h->poc_msb = h->prev_poc_msb + max_poc_lsb; | |
2836 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2) | |
2837 h->poc_msb = h->prev_poc_msb - max_poc_lsb; | |
2838 else | |
2839 h->poc_msb = h->prev_poc_msb; | |
2840 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb); | |
2841 field_poc[0] = | |
2842 field_poc[1] = h->poc_msb + h->poc_lsb; | |
2843 if(s->picture_structure == PICT_FRAME) | |
2844 field_poc[1] += h->delta_poc_bottom; | |
2845 }else if(h->sps.poc_type==1){ | |
2846 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc; | |
2847 int i; | |
2848 | |
2849 if(h->sps.poc_cycle_length != 0) | |
2850 abs_frame_num = h->frame_num_offset + h->frame_num; | |
2851 else | |
2852 abs_frame_num = 0; | |
2853 | |
2854 if(h->nal_ref_idc==0 && abs_frame_num > 0) | |
2855 abs_frame_num--; | |
2856 | |
2857 expected_delta_per_poc_cycle = 0; | |
2858 for(i=0; i < h->sps.poc_cycle_length; i++) | |
2859 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse | |
2860 | |
2861 if(abs_frame_num > 0){ | |
2862 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length; | |
2863 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length; | |
2864 | |
2865 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle; | |
2866 for(i = 0; i <= frame_num_in_poc_cycle; i++) | |
2867 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ]; | |
2868 } else | |
2869 expectedpoc = 0; | |
2870 | |
2871 if(h->nal_ref_idc == 0) | |
2872 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic; | |
2873 | |
2874 field_poc[0] = expectedpoc + h->delta_poc[0]; | |
2875 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field; | |
2876 | |
2877 if(s->picture_structure == PICT_FRAME) | |
2878 field_poc[1] += h->delta_poc[1]; | |
2879 }else{ | |
2880 int poc; | |
2881 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
2882 poc= 0; | |
2883 }else{ | |
2884 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num); | |
2885 else poc= 2*(h->frame_num_offset + h->frame_num) - 1; | |
2886 } | |
2887 field_poc[0]= poc; | |
2888 field_poc[1]= poc; | |
2889 } | |
2890 | |
2891 if(s->picture_structure != PICT_BOTTOM_FIELD) | |
2892 s->current_picture_ptr->field_poc[0]= field_poc[0]; | |
2893 if(s->picture_structure != PICT_TOP_FIELD) | |
2894 s->current_picture_ptr->field_poc[1]= field_poc[1]; | |
2895 if(s->picture_structure == PICT_FRAME) // FIXME field pix? | |
2896 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]); | |
2897 | |
2898 return 0; | |
2899 } | |
2900 | |
2901 /** | |
2902 * decodes a slice header. | |
2903 * this will allso call MPV_common_init() and frame_start() as needed | |
2904 */ | |
2905 static int decode_slice_header(H264Context *h){ | |
2906 MpegEncContext * const s = &h->s; | |
2907 int first_mb_in_slice, pps_id; | |
2908 int num_ref_idx_active_override_flag; | |
2909 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE}; | |
2910 | |
2911 s->current_picture.reference= h->nal_ref_idc != 0; | |
2912 | |
2913 first_mb_in_slice= get_ue_golomb(&s->gb); | |
2914 | |
2915 h->slice_type= get_ue_golomb(&s->gb); | |
2916 if(h->slice_type > 9){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2917 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 | 2918 } |
2919 if(h->slice_type > 4){ | |
2920 h->slice_type -= 5; | |
2921 h->slice_type_fixed=1; | |
2922 }else | |
2923 h->slice_type_fixed=0; | |
2924 | |
2925 h->slice_type= slice_type_map[ h->slice_type ]; | |
2926 | |
2927 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though | |
2928 | |
2929 pps_id= get_ue_golomb(&s->gb); | |
2930 if(pps_id>255){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
2931 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n"); |
1168 | 2932 return -1; |
2933 } | |
2934 h->pps= h->pps_buffer[pps_id]; | |
1174 | 2935 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
|
2936 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n"); |
1174 | 2937 return -1; |
2938 } | |
2939 | |
1168 | 2940 h->sps= h->sps_buffer[ h->pps.sps_id ]; |
1174 | 2941 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
|
2942 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n"); |
1174 | 2943 return -1; |
2944 } | |
1168 | 2945 |
2946 s->mb_width= h->sps.mb_width; | |
2947 s->mb_height= h->sps.mb_height; | |
2948 | |
2949 h->b_stride= s->mb_width*4; | |
2950 h->b8_stride= s->mb_width*2; | |
2951 | |
2952 s->mb_x = first_mb_in_slice % s->mb_width; | |
2953 s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW | |
2954 | |
1371 | 2955 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right ); |
1168 | 2956 if(h->sps.frame_mbs_only_flag) |
1371 | 2957 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom); |
1168 | 2958 else |
1371 | 2959 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck |
1168 | 2960 |
2961 if (s->context_initialized | |
1548 | 2962 && ( s->width != s->avctx->width || s->height != s->avctx->height)) { |
1168 | 2963 free_tables(h); |
2964 MPV_common_end(s); | |
2965 } | |
2966 if (!s->context_initialized) { | |
2967 if (MPV_common_init(s) < 0) | |
2968 return -1; | |
2969 | |
2970 alloc_tables(h); | |
2971 | |
2972 s->avctx->width = s->width; | |
2973 s->avctx->height = s->height; | |
1548 | 2974 s->avctx->sample_aspect_ratio= h->sps.sar; |
1168 | 2975 } |
2976 | |
2977 if(first_mb_in_slice == 0){ | |
2978 frame_start(h); | |
2979 } | |
2980 | |
1169 | 2981 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup |
1168 | 2982 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num); |
2983 | |
2984 if(h->sps.frame_mbs_only_flag){ | |
2985 s->picture_structure= PICT_FRAME; | |
2986 }else{ | |
2987 if(get_bits1(&s->gb)) //field_pic_flag | |
2988 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag | |
2989 else | |
2990 s->picture_structure= PICT_FRAME; | |
2991 } | |
2992 | |
2993 if(s->picture_structure==PICT_FRAME){ | |
2994 h->curr_pic_num= h->frame_num; | |
2995 h->max_pic_num= 1<< h->sps.log2_max_frame_num; | |
2996 }else{ | |
2997 h->curr_pic_num= 2*h->frame_num; | |
2998 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1); | |
2999 } | |
3000 | |
3001 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
1453 | 3002 get_ue_golomb(&s->gb); /* idr_pic_id */ |
1168 | 3003 } |
3004 | |
3005 if(h->sps.poc_type==0){ | |
3006 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb); | |
3007 | |
3008 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){ | |
3009 h->delta_poc_bottom= get_se_golomb(&s->gb); | |
3010 } | |
3011 } | |
3012 | |
3013 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){ | |
3014 h->delta_poc[0]= get_se_golomb(&s->gb); | |
3015 | |
3016 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME) | |
3017 h->delta_poc[1]= get_se_golomb(&s->gb); | |
3018 } | |
3019 | |
3020 init_poc(h); | |
3021 | |
3022 if(h->pps.redundant_pic_cnt_present){ | |
3023 h->redundant_pic_count= get_ue_golomb(&s->gb); | |
3024 } | |
3025 | |
3026 //set defaults, might be overriden a few line later | |
3027 h->ref_count[0]= h->pps.ref_count[0]; | |
3028 h->ref_count[1]= h->pps.ref_count[1]; | |
3029 | |
3030 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){ | |
3031 if(h->slice_type == B_TYPE){ | |
3032 h->direct_spatial_mv_pred= get_bits1(&s->gb); | |
3033 } | |
3034 num_ref_idx_active_override_flag= get_bits1(&s->gb); | |
3035 | |
3036 if(num_ref_idx_active_override_flag){ | |
3037 h->ref_count[0]= get_ue_golomb(&s->gb) + 1; | |
3038 if(h->slice_type==B_TYPE) | |
3039 h->ref_count[1]= get_ue_golomb(&s->gb) + 1; | |
3040 | |
3041 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
|
3042 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n"); |
1168 | 3043 return -1; |
3044 } | |
3045 } | |
3046 } | |
3047 | |
3048 if(first_mb_in_slice == 0){ | |
3049 fill_default_ref_list(h); | |
3050 } | |
3051 | |
3052 decode_ref_pic_list_reordering(h); | |
3053 | |
3054 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) | |
3055 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) ) | |
3056 pred_weight_table(h); | |
3057 | |
3058 if(s->current_picture.reference) | |
3059 decode_ref_pic_marking(h); | |
3060 //FIXME CABAC stuff | |
3061 | |
3062 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta | |
1898 | 3063 if(s->qscale<0 || s->qscale>51){ |
3064 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale); | |
3065 return -1; | |
3066 } | |
1168 | 3067 //FIXME qscale / qp ... stuff |
3068 if(h->slice_type == SP_TYPE){ | |
1453 | 3069 get_bits1(&s->gb); /* sp_for_switch_flag */ |
1168 | 3070 } |
3071 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){ | |
1453 | 3072 get_se_golomb(&s->gb); /* slice_qs_delta */ |
1168 | 3073 } |
3074 | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3075 h->deblocking_filter = 1; |
1898 | 3076 h->slice_alpha_c0_offset = 0; |
3077 h->slice_beta_offset = 0; | |
1168 | 3078 if( h->pps.deblocking_filter_parameters_present ) { |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3079 h->deblocking_filter= get_ue_golomb(&s->gb); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3080 if(h->deblocking_filter < 2) |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3081 h->deblocking_filter^= 1; // 1<->0 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3082 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3083 if( h->deblocking_filter ) { |
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
|
3084 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
|
3085 h->slice_beta_offset = get_se_golomb(&s->gb) << 1; |
1168 | 3086 } |
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
|
3087 } |
1168 | 3088 |
3089 #if 0 //FMO | |
3090 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5) | |
3091 slice_group_change_cycle= get_bits(&s->gb, ?); | |
3092 #endif | |
3093 | |
3094 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
|
3095 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 | 3096 first_mb_in_slice, |
1264 | 3097 av_get_pict_type_char(h->slice_type), |
1168 | 3098 pps_id, h->frame_num, |
3099 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1], | |
3100 h->ref_count[0], h->ref_count[1], | |
3101 s->qscale, | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3102 h->deblocking_filter |
1168 | 3103 ); |
3104 } | |
3105 | |
3106 return 0; | |
3107 } | |
3108 | |
3109 /** | |
3110 * | |
3111 */ | |
3112 static inline int get_level_prefix(GetBitContext *gb){ | |
3113 unsigned int buf; | |
3114 int log; | |
3115 | |
3116 OPEN_READER(re, gb); | |
3117 UPDATE_CACHE(re, gb); | |
3118 buf=GET_CACHE(re, gb); | |
3119 | |
3120 log= 32 - av_log2(buf); | |
3121 #ifdef TRACE | |
3122 print_bin(buf>>(32-log), log); | |
3123 printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__); | |
3124 #endif | |
3125 | |
3126 LAST_SKIP_BITS(re, gb, log); | |
3127 CLOSE_READER(re, gb); | |
3128 | |
3129 return log-1; | |
3130 } | |
3131 | |
3132 /** | |
3133 * decodes a residual block. | |
3134 * @param n block index | |
3135 * @param scantable scantable | |
3136 * @param max_coeff number of coefficients in the block | |
3137 * @return <0 if an error occured | |
3138 */ | |
3139 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){ | |
3140 MpegEncContext * const s = &h->s; | |
3141 const uint16_t *qmul= dequant_coeff[qp]; | |
3142 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}; | |
3143 int level[16], run[16]; | |
3144 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones; | |
3145 | |
3146 //FIXME put trailing_onex into the context | |
3147 | |
3148 if(n == CHROMA_DC_BLOCK_INDEX){ | |
3149 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1); | |
3150 total_coeff= coeff_token>>2; | |
3151 }else{ | |
3152 if(n == LUMA_DC_BLOCK_INDEX){ | |
3153 total_coeff= pred_non_zero_count(h, 0); | |
3154 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2); | |
3155 total_coeff= coeff_token>>2; | |
3156 }else{ | |
3157 total_coeff= pred_non_zero_count(h, n); | |
3158 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2); | |
3159 total_coeff= coeff_token>>2; | |
3160 h->non_zero_count_cache[ scan8[n] ]= total_coeff; | |
3161 } | |
3162 } | |
3163 | |
3164 //FIXME set last_non_zero? | |
3165 | |
3166 if(total_coeff==0) | |
3167 return 0; | |
3168 | |
3169 trailing_ones= coeff_token&3; | |
1170 | 3170 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff); |
1168 | 3171 assert(total_coeff<=16); |
3172 | |
3173 for(i=0; i<trailing_ones; i++){ | |
3174 level[i]= 1 - 2*get_bits1(gb); | |
3175 } | |
3176 | |
3177 suffix_length= total_coeff > 10 && trailing_ones < 3; | |
3178 | |
3179 for(; i<total_coeff; i++){ | |
3180 const int prefix= get_level_prefix(gb); | |
3181 int level_code, mask; | |
3182 | |
3183 if(prefix<14){ //FIXME try to build a large unified VLC table for all this | |
3184 if(suffix_length) | |
3185 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part | |
3186 else | |
3187 level_code= (prefix<<suffix_length); //part | |
3188 }else if(prefix==14){ | |
3189 if(suffix_length) | |
3190 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part | |
3191 else | |
3192 level_code= prefix + get_bits(gb, 4); //part | |
3193 }else if(prefix==15){ | |
3194 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part | |
3195 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense | |
3196 }else{ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3197 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y); |
1168 | 3198 return -1; |
3199 } | |
3200 | |
3201 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration | |
3202 | |
3203 mask= -(level_code&1); | |
3204 level[i]= (((2+level_code)>>1) ^ mask) - mask; | |
3205 | |
3206 if(suffix_length==0) suffix_length=1; //FIXME split first iteration | |
3207 | |
3208 #if 1 | |
3209 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++; | |
3210 #else | |
3211 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
|
3212 /* ? == prefix > 2 or sth */ |
1168 | 3213 #endif |
1170 | 3214 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length); |
1168 | 3215 } |
3216 | |
3217 if(total_coeff == max_coeff) | |
3218 zeros_left=0; | |
3219 else{ | |
3220 if(n == CHROMA_DC_BLOCK_INDEX) | |
3221 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1); | |
3222 else | |
3223 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1); | |
3224 } | |
3225 | |
3226 for(i=0; i<total_coeff-1; i++){ | |
3227 if(zeros_left <=0) | |
3228 break; | |
3229 else if(zeros_left < 7){ | |
3230 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1); | |
3231 }else{ | |
3232 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); | |
3233 } | |
3234 zeros_left -= run[i]; | |
3235 } | |
3236 | |
3237 if(zeros_left<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3238 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y); |
1168 | 3239 return -1; |
3240 } | |
3241 | |
3242 for(; i<total_coeff-1; i++){ | |
3243 run[i]= 0; | |
3244 } | |
3245 | |
3246 run[i]= zeros_left; | |
3247 | |
3248 coeff_num=-1; | |
3249 if(n > 24){ | |
3250 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode? | |
3251 int j; | |
3252 | |
3253 coeff_num += run[i] + 1; //FIXME add 1 earlier ? | |
3254 j= scantable[ coeff_num ]; | |
3255 | |
3256 block[j]= level[i]; | |
3257 } | |
3258 }else{ | |
3259 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode? | |
3260 int j; | |
3261 | |
3262 coeff_num += run[i] + 1; //FIXME add 1 earlier ? | |
3263 j= scantable[ coeff_num ]; | |
3264 | |
3265 block[j]= level[i] * qmul[j]; | |
3266 // printf("%d %d ", block[j], qmul[j]); | |
3267 } | |
3268 } | |
3269 return 0; | |
3270 } | |
3271 | |
3272 /** | |
3273 * decodes a macroblock | |
3274 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed | |
3275 */ | |
3276 static int decode_mb(H264Context *h){ | |
3277 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
|
3278 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1169 | 3279 int mb_type, partition_count, cbp; |
1168 | 3280 |
1252 | 3281 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong? |
1168 | 3282 |
1170 | 3283 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y); |
1453 | 3284 cbp = 0; /* avoid warning. FIXME: find a solution without slowing |
3285 down the code */ | |
1168 | 3286 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){ |
3287 if(s->mb_skip_run==-1) | |
3288 s->mb_skip_run= get_ue_golomb(&s->gb); | |
3289 | |
3290 if (s->mb_skip_run--) { | |
1187 | 3291 int mx, my; |
1168 | 3292 /* skip mb */ |
3293 //FIXME b frame | |
3294 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0; | |
3295 | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3296 memset(h->non_zero_count[mb_xy], 0, 16); |
1168 | 3297 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui |
3298 | |
3299 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){ | |
3300 h->mb_field_decoding_flag= get_bits1(&s->gb); | |
3301 } | |
3302 | |
3303 if(h->mb_field_decoding_flag) | |
3304 mb_type|= MB_TYPE_INTERLACED; | |
3305 | |
3306 fill_caches(h, mb_type); //FIXME check what is needed and what not ... | |
3307 pred_pskip_motion(h, &mx, &my); | |
3308 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1); | |
1269 | 3309 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4); |
1168 | 3310 write_back_motion(h, mb_type); |
3311 | |
3312 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
|
3313 s->current_picture.qscale_table[mb_xy]= s->qscale; |
1168 | 3314 h->slice_table[ mb_xy ]= h->slice_num; |
3315 | |
3316 h->prev_mb_skiped= 1; | |
3317 return 0; | |
3318 } | |
3319 } | |
3320 if(h->sps.mb_aff /* && !field pic FIXME needed? */){ | |
3321 if((s->mb_y&1)==0) | |
3322 h->mb_field_decoding_flag = get_bits1(&s->gb); | |
3323 }else | |
3324 h->mb_field_decoding_flag=0; //FIXME som ed note ?! | |
3325 | |
3326 h->prev_mb_skiped= 0; | |
3327 | |
3328 mb_type= get_ue_golomb(&s->gb); | |
3329 if(h->slice_type == B_TYPE){ | |
3330 if(mb_type < 23){ | |
3331 partition_count= b_mb_type_info[mb_type].partition_count; | |
3332 mb_type= b_mb_type_info[mb_type].type; | |
3333 }else{ | |
3334 mb_type -= 23; | |
3335 goto decode_intra_mb; | |
3336 } | |
3337 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){ | |
3338 if(mb_type < 5){ | |
3339 partition_count= p_mb_type_info[mb_type].partition_count; | |
3340 mb_type= p_mb_type_info[mb_type].type; | |
3341 }else{ | |
3342 mb_type -= 5; | |
3343 goto decode_intra_mb; | |
3344 } | |
3345 }else{ | |
3346 assert(h->slice_type == I_TYPE); | |
3347 decode_intra_mb: | |
3348 if(mb_type > 25){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3349 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 | 3350 return -1; |
3351 } | |
3352 partition_count=0; | |
3353 cbp= i_mb_type_info[mb_type].cbp; | |
3354 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode; | |
3355 mb_type= i_mb_type_info[mb_type].type; | |
3356 } | |
3357 | |
3358 if(h->mb_field_decoding_flag) | |
3359 mb_type |= MB_TYPE_INTERLACED; | |
3360 | |
3361 s->current_picture.mb_type[mb_xy]= mb_type; | |
3362 h->slice_table[ mb_xy ]= h->slice_num; | |
3363 | |
3364 if(IS_INTRA_PCM(mb_type)){ | |
3365 const uint8_t *ptr; | |
1187 | 3366 int x, y; |
1168 | 3367 |
3368 // we assume these blocks are very rare so we dont optimize it | |
3369 align_get_bits(&s->gb); | |
3370 | |
3371 ptr= s->gb.buffer + get_bits_count(&s->gb); | |
3372 | |
3373 for(y=0; y<16; y++){ | |
3374 const int index= 4*(y&3) + 64*(y>>2); | |
3375 for(x=0; x<16; x++){ | |
3376 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++); | |
3377 } | |
3378 } | |
3379 for(y=0; y<8; y++){ | |
3380 const int index= 256 + 4*(y&3) + 32*(y>>2); | |
3381 for(x=0; x<8; x++){ | |
3382 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++); | |
3383 } | |
3384 } | |
3385 for(y=0; y<8; y++){ | |
3386 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2); | |
3387 for(x=0; x<8; x++){ | |
3388 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++); | |
3389 } | |
3390 } | |
3391 | |
3392 skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers | |
3393 | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3394 //FIXME deblock filter, non_zero_count_cache init ... |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3395 memset(h->non_zero_count[mb_xy], 16, 16); |
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
|
3396 s->current_picture.qscale_table[mb_xy]= s->qscale; |
1168 | 3397 |
3398 return 0; | |
3399 } | |
3400 | |
3401 fill_caches(h, mb_type); | |
3402 | |
3403 //mb_pred | |
3404 if(IS_INTRA(mb_type)){ | |
3405 // init_top_left_availability(h); | |
3406 if(IS_INTRA4x4(mb_type)){ | |
3407 int i; | |
3408 | |
3409 // fill_intra4x4_pred_table(h); | |
3410 for(i=0; i<16; i++){ | |
3411 const int mode_coded= !get_bits1(&s->gb); | |
3412 const int predicted_mode= pred_intra_mode(h, i); | |
3413 int mode; | |
3414 | |
3415 if(mode_coded){ | |
3416 const int rem_mode= get_bits(&s->gb, 3); | |
3417 if(rem_mode<predicted_mode) | |
3418 mode= rem_mode; | |
3419 else | |
3420 mode= rem_mode + 1; | |
3421 }else{ | |
3422 mode= predicted_mode; | |
3423 } | |
3424 | |
3425 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode; | |
3426 } | |
3427 write_back_intra_pred_mode(h); | |
3428 if( check_intra4x4_pred_mode(h) < 0) | |
3429 return -1; | |
3430 }else{ | |
3431 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode); | |
3432 if(h->intra16x16_pred_mode < 0) | |
3433 return -1; | |
3434 } | |
3435 h->chroma_pred_mode= get_ue_golomb(&s->gb); | |
3436 | |
3437 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode); | |
3438 if(h->chroma_pred_mode < 0) | |
3439 return -1; | |
3440 }else if(partition_count==4){ | |
3441 int i, j, sub_partition_count[4], list, ref[2][4]; | |
3442 | |
3443 if(h->slice_type == B_TYPE){ | |
3444 for(i=0; i<4; i++){ | |
3445 h->sub_mb_type[i]= get_ue_golomb(&s->gb); | |
3446 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
|
3447 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 | 3448 return -1; |
3449 } | |
3450 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; | |
3451 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type; | |
3452 } | |
3453 }else{ | |
3454 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ? | |
3455 for(i=0; i<4; i++){ | |
3456 h->sub_mb_type[i]= get_ue_golomb(&s->gb); | |
3457 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
|
3458 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 | 3459 return -1; |
3460 } | |
3461 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; | |
3462 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type; | |
3463 } | |
3464 } | |
3465 | |
3466 for(list=0; list<2; list++){ | |
3467 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list]; | |
3468 if(ref_count == 0) continue; | |
3469 for(i=0; i<4; i++){ | |
3470 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){ | |
3471 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip? | |
3472 }else{ | |
3473 //FIXME | |
3474 ref[list][i] = -1; | |
3475 } | |
3476 } | |
3477 } | |
3478 | |
3479 for(list=0; list<2; list++){ | |
3480 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list]; | |
3481 if(ref_count == 0) continue; | |
3482 | |
3483 for(i=0; i<4; i++){ | |
3484 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]= | |
3485 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i]; | |
3486 | |
3487 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){ | |
3488 const int sub_mb_type= h->sub_mb_type[i]; | |
3489 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1; | |
3490 for(j=0; j<sub_partition_count[i]; j++){ | |
3491 int mx, my; | |
3492 const int index= 4*i + block_width*j; | |
3493 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ]; | |
3494 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my); | |
3495 mx += get_se_golomb(&s->gb); | |
3496 my += get_se_golomb(&s->gb); | |
1170 | 3497 tprintf("final mv:%d %d\n", mx, my); |
3498 | |
1168 | 3499 if(IS_SUB_8X8(sub_mb_type)){ |
3500 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= | |
3501 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx; | |
3502 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= | |
3503 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my; | |
3504 }else if(IS_SUB_8X4(sub_mb_type)){ | |
3505 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx; | |
3506 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my; | |
3507 }else if(IS_SUB_4X8(sub_mb_type)){ | |
3508 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx; | |
3509 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my; | |
3510 }else{ | |
3511 assert(IS_SUB_4X4(sub_mb_type)); | |
3512 mv_cache[ 0 ][0]= mx; | |
3513 mv_cache[ 0 ][1]= my; | |
3514 } | |
3515 } | |
3516 }else{ | |
3517 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0]; | |
3518 p[0] = p[1]= | |
3519 p[8] = p[9]= 0; | |
3520 } | |
3521 } | |
3522 } | |
3523 }else if(!IS_DIRECT(mb_type)){ | |
3524 int list, mx, my, i; | |
3525 //FIXME we should set ref_idx_l? to 0 if we use that later ... | |
3526 if(IS_16X16(mb_type)){ | |
3527 for(list=0; list<2; list++){ | |
3528 if(h->ref_count[0]>0){ | |
3529 if(IS_DIR(mb_type, 0, list)){ | |
3530 const int val= get_te0_golomb(&s->gb, h->ref_count[list]); | |
3531 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1); | |
3532 } | |
3533 } | |
3534 } | |
3535 for(list=0; list<2; list++){ | |
3536 if(IS_DIR(mb_type, 0, list)){ | |
3537 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my); | |
3538 mx += get_se_golomb(&s->gb); | |
3539 my += get_se_golomb(&s->gb); | |
1170 | 3540 tprintf("final mv:%d %d\n", mx, my); |
3541 | |
1269 | 3542 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4); |
1168 | 3543 } |
3544 } | |
3545 } | |
3546 else if(IS_16X8(mb_type)){ | |
3547 for(list=0; list<2; list++){ | |
3548 if(h->ref_count[list]>0){ | |
3549 for(i=0; i<2; i++){ | |
3550 if(IS_DIR(mb_type, i, list)){ | |
3551 const int val= get_te0_golomb(&s->gb, h->ref_count[list]); | |
3552 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1); | |
3553 } | |
3554 } | |
3555 } | |
3556 } | |
3557 for(list=0; list<2; list++){ | |
3558 for(i=0; i<2; i++){ | |
3559 if(IS_DIR(mb_type, i, list)){ | |
3560 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my); | |
3561 mx += get_se_golomb(&s->gb); | |
3562 my += get_se_golomb(&s->gb); | |
1170 | 3563 tprintf("final mv:%d %d\n", mx, my); |
3564 | |
1269 | 3565 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4); |
1168 | 3566 } |
3567 } | |
3568 } | |
3569 }else{ | |
3570 assert(IS_8X16(mb_type)); | |
3571 for(list=0; list<2; list++){ | |
3572 if(h->ref_count[list]>0){ | |
3573 for(i=0; i<2; i++){ | |
3574 if(IS_DIR(mb_type, i, list)){ //FIXME optimize | |
3575 const int val= get_te0_golomb(&s->gb, h->ref_count[list]); | |
3576 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1); | |
3577 } | |
3578 } | |
3579 } | |
3580 } | |
3581 for(list=0; list<2; list++){ | |
3582 for(i=0; i<2; i++){ | |
3583 if(IS_DIR(mb_type, i, list)){ | |
3584 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my); | |
3585 mx += get_se_golomb(&s->gb); | |
3586 my += get_se_golomb(&s->gb); | |
1170 | 3587 tprintf("final mv:%d %d\n", mx, my); |
3588 | |
1269 | 3589 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4); |
1168 | 3590 } |
3591 } | |
3592 } | |
3593 } | |
3594 } | |
3595 | |
3596 if(IS_INTER(mb_type)) | |
3597 write_back_motion(h, mb_type); | |
3598 | |
3599 if(!IS_INTRA16x16(mb_type)){ | |
3600 cbp= get_ue_golomb(&s->gb); | |
3601 if(cbp > 47){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3602 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y); |
1168 | 3603 return -1; |
3604 } | |
3605 | |
3606 if(IS_INTRA4x4(mb_type)) | |
3607 cbp= golomb_to_intra4x4_cbp[cbp]; | |
3608 else | |
3609 cbp= golomb_to_inter_cbp[cbp]; | |
3610 } | |
3611 | |
3612 if(cbp || IS_INTRA16x16(mb_type)){ | |
3613 int i8x8, i4x4, chroma_idx; | |
3614 int chroma_qp, dquant; | |
3615 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr; | |
3616 const uint8_t *scan, *dc_scan; | |
3617 | |
3618 // fill_non_zero_count_cache(h); | |
3619 | |
3620 if(IS_INTERLACED(mb_type)){ | |
3621 scan= field_scan; | |
3622 dc_scan= luma_dc_field_scan; | |
3623 }else{ | |
3624 scan= zigzag_scan; | |
3625 dc_scan= luma_dc_zigzag_scan; | |
3626 } | |
3627 | |
3628 dquant= get_se_golomb(&s->gb); | |
3629 | |
3630 if( dquant > 25 || dquant < -26 ){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
3631 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 | 3632 return -1; |
3633 } | |
3634 | |
3635 s->qscale += dquant; | |
3636 if(((unsigned)s->qscale) > 51){ | |
3637 if(s->qscale<0) s->qscale+= 52; | |
3638 else s->qscale-= 52; | |
3639 } | |
3640 | |
3641 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale); | |
3642 if(IS_INTRA16x16(mb_type)){ | |
3643 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){ | |
3644 return -1; //FIXME continue if partotioned and other retirn -1 too | |
3645 } | |
3646 | |
3647 assert((cbp&15) == 0 || (cbp&15) == 15); | |
3648 | |
3649 if(cbp&15){ | |
3650 for(i8x8=0; i8x8<4; i8x8++){ | |
3651 for(i4x4=0; i4x4<4; i4x4++){ | |
3652 const int index= i4x4 + 4*i8x8; | |
3653 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){ | |
3654 return -1; | |
3655 } | |
3656 } | |
3657 } | |
3658 }else{ | |
1636 | 3659 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1); |
1168 | 3660 } |
3661 }else{ | |
3662 for(i8x8=0; i8x8<4; i8x8++){ | |
3663 if(cbp & (1<<i8x8)){ | |
3664 for(i4x4=0; i4x4<4; i4x4++){ | |
3665 const int index= i4x4 + 4*i8x8; | |
3666 | |
3667 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){ | |
3668 return -1; | |
3669 } | |
3670 } | |
3671 }else{ | |
3672 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ]; | |
3673 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0; | |
3674 } | |
3675 } | |
3676 } | |
3677 | |
3678 if(cbp&0x30){ | |
3679 for(chroma_idx=0; chroma_idx<2; chroma_idx++) | |
3680 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){ | |
3681 return -1; | |
3682 } | |
3683 } | |
3684 | |
3685 if(cbp&0x20){ | |
3686 for(chroma_idx=0; chroma_idx<2; chroma_idx++){ | |
3687 for(i4x4=0; i4x4<4; i4x4++){ | |
3688 const int index= 16 + 4*chroma_idx + i4x4; | |
3689 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){ | |
3690 return -1; | |
3691 } | |
3692 } | |
3693 } | |
3694 }else{ | |
3695 uint8_t * const nnz= &h->non_zero_count_cache[0]; | |
3696 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] = | |
3697 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0; | |
3698 } | |
3699 }else{ | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3700 uint8_t * const nnz= &h->non_zero_count_cache[0]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3701 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3702 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] = |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3703 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0; |
1168 | 3704 } |
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
|
3705 s->current_picture.qscale_table[mb_xy]= s->qscale; |
1168 | 3706 write_back_non_zero_count(h); |
3707 | |
3708 return 0; | |
3709 } | |
3710 | |
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
|
3711 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
|
3712 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
|
3713 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
|
3714 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
|
3715 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
|
3716 |
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 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
|
3718 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
|
3719 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
|
3720 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
|
3721 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3722 |
1898 | 3723 if( bS[i] < 4 ) { |
3724 const int tc0 = tc0_table[index_a][bS[i] - 1]; | |
3725 /* 4px edge length */ | |
3726 for( d = 0; d < 4; d++ ) { | |
3727 const int p0 = pix[-1]; | |
3728 const int p1 = pix[-2]; | |
3729 const int p2 = pix[-3]; | |
3730 const int q0 = pix[0]; | |
3731 const int q1 = pix[1]; | |
3732 const int q2 = pix[2]; | |
3733 | |
3734 if( ABS( p0 - q0 ) < alpha && | |
3735 ABS( p1 - p0 ) < beta && | |
3736 ABS( q1 - q0 ) < beta ) { | |
3737 int tc = tc0; | |
3738 int i_delta; | |
3739 | |
3740 if( ABS( p2 - p0 ) < beta ) { | |
3741 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 ); | |
3742 tc++; | |
3743 } | |
3744 if( ABS( q2 - q0 ) < beta ) { | |
3745 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 ); | |
3746 tc++; | |
3747 } | |
3748 | |
3749 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); | |
3750 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */ | |
3751 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
|
3752 } |
1898 | 3753 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
|
3754 } |
1898 | 3755 }else{ |
3756 /* 4px edge length */ | |
3757 for( d = 0; d < 4; d++ ) { | |
3758 const int p0 = pix[-1]; | |
3759 const int p1 = pix[-2]; | |
3760 const int p2 = pix[-3]; | |
3761 | |
3762 const int q0 = pix[0]; | |
3763 const int q1 = pix[1]; | |
3764 const int q2 = pix[2]; | |
3765 | |
3766 if( ABS( p0 - q0 ) < alpha && | |
3767 ABS( p1 - p0 ) < beta && | |
3768 ABS( q1 - q0 ) < beta ) { | |
3769 | |
3770 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){ | |
3771 if( ABS( p2 - p0 ) < beta) | |
3772 { | |
3773 const int p3 = pix[-4]; | |
3774 /* p0', p1', p2' */ | |
3775 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3; | |
3776 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2; | |
3777 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3; | |
3778 } else { | |
3779 /* p0' */ | |
3780 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
3781 } | |
3782 if( ABS( q2 - q0 ) < beta) | |
3783 { | |
3784 const int q3 = pix[3]; | |
3785 /* q0', q1', q2' */ | |
3786 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3; | |
3787 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2; | |
3788 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3; | |
3789 } else { | |
3790 /* q0' */ | |
3791 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
3792 } | |
3793 }else{ | |
3794 /* p0', q0' */ | |
3795 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
3796 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
3797 } | |
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
|
3798 } |
1898 | 3799 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
|
3800 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3801 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3802 } |
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 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3804 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
|
3805 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
|
3806 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
|
3807 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
|
3808 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
|
3809 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3810 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
|
3811 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
|
3812 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
|
3813 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
|
3814 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3815 |
1898 | 3816 if( bS[i] < 4 ) { |
3817 const int tc = tc0_table[index_a][bS[i] - 1] + 1; | |
3818 /* 2px edge length (because we use same bS than the one for luma) */ | |
3819 for( d = 0; d < 2; d++ ){ | |
3820 const int p0 = pix[-1]; | |
3821 const int p1 = pix[-2]; | |
3822 const int q0 = pix[0]; | |
3823 const int q1 = pix[1]; | |
3824 | |
3825 if( ABS( p0 - q0 ) < alpha && | |
3826 ABS( p1 - p0 ) < beta && | |
3827 ABS( q1 - q0 ) < beta ) { | |
3828 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); | |
3829 | |
3830 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */ | |
3831 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */ | |
3832 } | |
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
|
3833 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
|
3834 } |
1898 | 3835 }else{ |
3836 /* 2px edge length (because we use same bS than the one for luma) */ | |
3837 for( d = 0; d < 2; d++ ){ | |
3838 const int p0 = pix[-1]; | |
3839 const int p1 = pix[-2]; | |
3840 const int q0 = pix[0]; | |
3841 const int q1 = pix[1]; | |
3842 | |
3843 if( ABS( p0 - q0 ) < alpha && | |
3844 ABS( p1 - p0 ) < beta && | |
3845 ABS( q1 - q0 ) < beta ) { | |
3846 | |
3847 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */ | |
3848 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */ | |
3849 } | |
3850 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
|
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 } |
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 |
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 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
|
3857 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
|
3858 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
|
3859 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
|
3860 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
|
3861 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
|
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 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
|
3864 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
|
3865 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
|
3866 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
|
3867 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
3868 |
1898 | 3869 if( bS[i] < 4 ) { |
3870 const int tc0 = tc0_table[index_a][bS[i] - 1]; | |
3871 /* 4px edge length */ | |
3872 for( d = 0; d < 4; d++ ) { | |
3873 const int p0 = pix[-1*pix_next]; | |
3874 const int p1 = pix[-2*pix_next]; | |
3875 const int p2 = pix[-3*pix_next]; | |
3876 const int q0 = pix[0]; | |
3877 const int q1 = pix[1*pix_next]; | |
3878 const int q2 = pix[2*pix_next]; | |
3879 | |
3880 if( ABS( p0 - q0 ) < alpha && | |
3881 ABS( p1 - p0 ) < beta && | |
3882 ABS( q1 - q0 ) < beta ) { | |
3883 | |
3884 int tc = tc0; | |
3885 int i_delta; | |
3886 | |
3887 if( ABS( p2 - p0 ) < beta ) { | |
3888 pix[-2*pix_next] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 ); | |
3889 tc++; | |
3890 } | |
3891 if( ABS( q2 - q0 ) < beta ) { | |
3892 pix[pix_next] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 ); | |
3893 tc++; | |
3894 } | |
3895 | |
3896 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); | |
3897 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */ | |
3898 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
|
3899 } |
1898 | 3900 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
|
3901 } |
1898 | 3902 }else{ |
3903 /* 4px edge length */ | |
3904 for( d = 0; d < 4; d++ ) { | |
3905 const int p0 = pix[-1*pix_next]; | |
3906 const int p1 = pix[-2*pix_next]; | |
3907 const int p2 = pix[-3*pix_next]; | |
3908 const int q0 = pix[0]; | |
3909 const int q1 = pix[1*pix_next]; | |
3910 const int q2 = pix[2*pix_next]; | |
3911 | |
3912 if( ABS( p0 - q0 ) < alpha && | |
3913 ABS( p1 - p0 ) < beta && | |
3914 ABS( q1 - q0 ) < beta ) { | |
3915 | |
3916 const int p3 = pix[-4*pix_next]; | |
3917 const int q3 = pix[ 3*pix_next]; | |
3918 | |
3919 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){ | |
3920 if( ABS( p2 - p0 ) < beta) { | |
3921 /* p0', p1', p2' */ | |
3922 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3; | |
3923 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2; | |
3924 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3; | |
3925 } else { | |
3926 /* p0' */ | |
3927 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
3928 } | |
3929 if( ABS( q2 - q0 ) < beta) { | |
3930 /* q0', q1', q2' */ | |
3931 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3; | |
3932 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2; | |
3933 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3; | |
3934 } else { | |
3935 /* q0' */ | |
3936 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
3937 } | |
3938 }else{ | |
3939 /* p0', q0' */ | |
3940 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
3941 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
3942 } | |
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
|
3943 } |
1898 | 3944 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
|
3945 } |
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 } |
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 } |
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 } |
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 |
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 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
|
3951 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
|
3952 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
|
3953 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
|
3954 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
|
3955 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
|
3956 |
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 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
|
3958 { |
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 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
|
3960 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
|
3961 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
|
3962 } |
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 |
1898 | 3964 if( bS[i] < 4 ) { |
3965 int tc = tc0_table[index_a][bS[i] - 1] + 1; | |
3966 /* 2px edge length (see deblocking_filter_edgecv) */ | |
3967 for( d = 0; d < 2; d++ ) { | |
3968 const int p0 = pix[-1*pix_next]; | |
3969 const int p1 = pix[-2*pix_next]; | |
3970 const int q0 = pix[0]; | |
3971 const int q1 = pix[1*pix_next]; | |
3972 | |
3973 if( ABS( p0 - q0 ) < alpha && | |
3974 ABS( p1 - p0 ) < beta && | |
3975 ABS( q1 - q0 ) < beta ) { | |
3976 | |
3977 int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); | |
3978 | |
3979 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */ | |
3980 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */ | |
3981 } | |
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
|
3982 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
|
3983 } |
1898 | 3984 }else{ |
3985 /* 2px edge length (see deblocking_filter_edgecv) */ | |
3986 for( d = 0; d < 2; d++ ) { | |
3987 const int p0 = pix[-1*pix_next]; | |
3988 const int p1 = pix[-2*pix_next]; | |
3989 const int q0 = pix[0]; | |
3990 const int q1 = pix[1*pix_next]; | |
3991 | |
3992 if( ABS( p0 - q0 ) < alpha && | |
3993 ABS( p1 - p0 ) < beta && | |
3994 ABS( q1 - q0 ) < beta ) { | |
3995 | |
3996 pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */ | |
3997 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */ | |
3998 } | |
3999 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
|
4000 } |
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 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4002 } |
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 } |
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 |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4005 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr) { |
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
|
4006 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
|
4007 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
|
4008 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
|
4009 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
|
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 /* 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
|
4012 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
|
4013 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
|
4014 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4015 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
|
4016 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
|
4017 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4018 /* 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
|
4019 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
|
4020 { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4021 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
|
4022 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
|
4023 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4024 /* 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
|
4025 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
|
4026 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
|
4027 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4028 /* FIXME test slice boundary */ |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4029 if( h->deblocking_filter == 2 ) { |
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
|
4030 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4031 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4032 /* 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
|
4033 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
|
4034 /* 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
|
4035 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
|
4036 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
|
4037 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
|
4038 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4039 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
|
4040 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
|
4041 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
|
4042 } 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
|
4043 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
|
4044 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
|
4045 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
|
4046 int y = dir == 0 ? i : edge; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4047 int b_idx= 8 + 4 + x + 8*y; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4048 int bn_idx= b_idx - (dir ? 8:1); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4049 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4050 if( h->non_zero_count_cache[b_idx] != 0 || |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4051 h->non_zero_count_cache[bn_idx] != 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
|
4052 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
|
4053 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4054 else if( h->slice_type == P_TYPE ) { |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4055 if( h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx] || |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4056 ABS( h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] ) >= 4 || |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4057 ABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][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
|
4058 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
|
4059 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
|
4060 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
|
4061 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4062 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
|
4063 /* 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
|
4064 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
|
4065 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4066 } |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4067 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4068 if(bS[0]+bS[1]+bS[2]+bS[3] == 0) |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4069 continue; |
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
|
4070 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4071 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4072 /* Filter edge */ |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4073 qp = ( s->qscale + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 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
|
4074 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
|
4075 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp ); |
1898 | 4076 if( (edge&1) == 0 ) { |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4077 int chroma_qp = ( h->chroma_qp + |
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
|
4078 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
|
4079 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
|
4080 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
|
4081 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4082 } 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
|
4083 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp ); |
1898 | 4084 if( (edge&1) == 0 ) { |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4085 int chroma_qp = ( h->chroma_qp + |
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
|
4086 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
|
4087 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
|
4088 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
|
4089 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4090 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4091 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4092 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4093 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4094 |
1168 | 4095 static int decode_slice(H264Context *h){ |
4096 MpegEncContext * const s = &h->s; | |
4097 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F; | |
4098 | |
4099 s->mb_skip_run= -1; | |
4100 | |
4101 #if 1 | |
4102 for(;;){ | |
4103 int ret= decode_mb(h); | |
4104 | |
4105 hl_decode_mb(h); | |
4106 | |
4107 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ? | |
4108 s->mb_y++; | |
4109 ret= decode_mb(h); | |
4110 | |
4111 hl_decode_mb(h); | |
4112 s->mb_y--; | |
4113 } | |
4114 | |
4115 if(ret<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4116 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y); |
1168 | 4117 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); |
4118 | |
4119 return -1; | |
4120 } | |
4121 | |
4122 if(++s->mb_x >= s->mb_width){ | |
4123 s->mb_x=0; | |
4124 ff_draw_horiz_band(s, 16*s->mb_y, 16); | |
4125 if(++s->mb_y >= s->mb_height){ | |
1170 | 4126 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits); |
4127 | |
1168 | 4128 if(get_bits_count(&s->gb) == s->gb.size_in_bits){ |
4129 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); | |
4130 | |
4131 return 0; | |
4132 }else{ | |
4133 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); | |
4134 | |
4135 return -1; | |
4136 } | |
4137 } | |
4138 } | |
4139 | |
4140 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){ | |
4141 if(get_bits_count(&s->gb) == s->gb.size_in_bits){ | |
4142 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); | |
4143 | |
4144 return 0; | |
4145 }else{ | |
4146 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); | |
4147 | |
4148 return -1; | |
4149 } | |
4150 } | |
4151 } | |
4152 #endif | |
4153 #if 0 | |
4154 for(;s->mb_y < s->mb_height; s->mb_y++){ | |
4155 for(;s->mb_x < s->mb_width; s->mb_x++){ | |
4156 int ret= decode_mb(h); | |
4157 | |
4158 hl_decode_mb(h); | |
4159 | |
4160 if(ret<0){ | |
4161 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y); | |
4162 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); | |
4163 | |
4164 return -1; | |
4165 } | |
4166 | |
4167 if(++s->mb_x >= s->mb_width){ | |
4168 s->mb_x=0; | |
4169 if(++s->mb_y >= s->mb_height){ | |
4170 if(get_bits_count(s->gb) == s->gb.size_in_bits){ | |
4171 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); | |
4172 | |
4173 return 0; | |
4174 }else{ | |
4175 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); | |
4176 | |
4177 return -1; | |
4178 } | |
4179 } | |
4180 } | |
4181 | |
4182 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){ | |
4183 if(get_bits_count(s->gb) == s->gb.size_in_bits){ | |
4184 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); | |
4185 | |
4186 return 0; | |
4187 }else{ | |
4188 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); | |
4189 | |
4190 return -1; | |
4191 } | |
4192 } | |
4193 } | |
4194 s->mb_x=0; | |
4195 ff_draw_horiz_band(s, 16*s->mb_y, 16); | |
4196 } | |
4197 #endif | |
4198 return -1; //not reached | |
4199 } | |
4200 | |
4201 static inline int decode_vui_parameters(H264Context *h, SPS *sps){ | |
4202 MpegEncContext * const s = &h->s; | |
4203 int aspect_ratio_info_present_flag, aspect_ratio_idc; | |
4204 | |
4205 aspect_ratio_info_present_flag= get_bits1(&s->gb); | |
4206 | |
4207 if( aspect_ratio_info_present_flag ) { | |
4208 aspect_ratio_idc= get_bits(&s->gb, 8); | |
4209 if( aspect_ratio_idc == EXTENDED_SAR ) { | |
1548 | 4210 sps->sar.num= get_bits(&s->gb, 16); |
4211 sps->sar.den= get_bits(&s->gb, 16); | |
1168 | 4212 }else if(aspect_ratio_idc < 16){ |
1548 | 4213 sps->sar= pixel_aspect[aspect_ratio_idc]; |
1168 | 4214 }else{ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4215 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n"); |
1168 | 4216 return -1; |
4217 } | |
4218 }else{ | |
1548 | 4219 sps->sar.num= |
4220 sps->sar.den= 0; | |
1168 | 4221 } |
4222 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height); | |
4223 #if 0 | |
4224 | overscan_info_present_flag |0 |u(1) | | |
4225 | if( overscan_info_present_flag ) | | | | |
4226 | overscan_appropriate_flag |0 |u(1) | | |
4227 | video_signal_type_present_flag |0 |u(1) | | |
4228 | if( video_signal_type_present_flag ) { | | | | |
4229 | video_format |0 |u(3) | | |
4230 | video_full_range_flag |0 |u(1) | | |
4231 | colour_description_present_flag |0 |u(1) | | |
4232 | if( colour_description_present_flag ) { | | | | |
4233 | colour_primaries |0 |u(8) | | |
4234 | transfer_characteristics |0 |u(8) | | |
4235 | matrix_coefficients |0 |u(8) | | |
4236 | } | | | | |
4237 | } | | | | |
4238 | chroma_location_info_present_flag |0 |u(1) | | |
4239 | if ( chroma_location_info_present_flag ) { | | | | |
4240 | chroma_sample_location_type_top_field |0 |ue(v) | | |
4241 | chroma_sample_location_type_bottom_field |0 |ue(v) | | |
4242 | } | | | | |
4243 | timing_info_present_flag |0 |u(1) | | |
4244 | if( timing_info_present_flag ) { | | | | |
4245 | num_units_in_tick |0 |u(32) | | |
4246 | time_scale |0 |u(32) | | |
4247 | fixed_frame_rate_flag |0 |u(1) | | |
4248 | } | | | | |
4249 | nal_hrd_parameters_present_flag |0 |u(1) | | |
4250 | if( nal_hrd_parameters_present_flag = = 1) | | | | |
4251 | hrd_parameters( ) | | | | |
4252 | vcl_hrd_parameters_present_flag |0 |u(1) | | |
4253 | if( vcl_hrd_parameters_present_flag = = 1) | | | | |
4254 | hrd_parameters( ) | | | | |
4255 | if( ( nal_hrd_parameters_present_flag = = 1 | || | | | |
4256 | | | | | |
4257 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | | | |
4258 | low_delay_hrd_flag |0 |u(1) | | |
4259 | bitstream_restriction_flag |0 |u(1) | | |
4260 | if( bitstream_restriction_flag ) { |0 |u(1) | | |
4261 | motion_vectors_over_pic_boundaries_flag |0 |u(1) | | |
4262 | max_bytes_per_pic_denom |0 |ue(v) | | |
4263 | max_bits_per_mb_denom |0 |ue(v) | | |
4264 | log2_max_mv_length_horizontal |0 |ue(v) | | |
4265 | log2_max_mv_length_vertical |0 |ue(v) | | |
4266 | num_reorder_frames |0 |ue(v) | | |
4267 | max_dec_frame_buffering |0 |ue(v) | | |
4268 | } | | | | |
4269 |} | | | | |
4270 #endif | |
4271 return 0; | |
4272 } | |
4273 | |
4274 static inline int decode_seq_parameter_set(H264Context *h){ | |
4275 MpegEncContext * const s = &h->s; | |
1371 | 4276 int profile_idc, level_idc; |
1168 | 4277 int sps_id, i; |
4278 SPS *sps; | |
4279 | |
4280 profile_idc= get_bits(&s->gb, 8); | |
1371 | 4281 get_bits1(&s->gb); //constraint_set0_flag |
4282 get_bits1(&s->gb); //constraint_set1_flag | |
4283 get_bits1(&s->gb); //constraint_set2_flag | |
4284 get_bits(&s->gb, 5); // reserved | |
1168 | 4285 level_idc= get_bits(&s->gb, 8); |
4286 sps_id= get_ue_golomb(&s->gb); | |
4287 | |
4288 sps= &h->sps_buffer[ sps_id ]; | |
4289 sps->profile_idc= profile_idc; | |
4290 sps->level_idc= level_idc; | |
4291 | |
4292 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4; | |
4293 sps->poc_type= get_ue_golomb(&s->gb); | |
4294 | |
4295 if(sps->poc_type == 0){ //FIXME #define | |
4296 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4; | |
4297 } else if(sps->poc_type == 1){//FIXME #define | |
4298 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb); | |
4299 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb); | |
4300 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb); | |
4301 sps->poc_cycle_length= get_ue_golomb(&s->gb); | |
4302 | |
4303 for(i=0; i<sps->poc_cycle_length; i++) | |
4304 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb); | |
4305 } | |
4306 if(sps->poc_type > 2){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4307 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type); |
1168 | 4308 return -1; |
4309 } | |
4310 | |
4311 sps->ref_frame_count= get_ue_golomb(&s->gb); | |
1371 | 4312 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb); |
1168 | 4313 sps->mb_width= get_ue_golomb(&s->gb) + 1; |
4314 sps->mb_height= get_ue_golomb(&s->gb) + 1; | |
4315 sps->frame_mbs_only_flag= get_bits1(&s->gb); | |
4316 if(!sps->frame_mbs_only_flag) | |
4317 sps->mb_aff= get_bits1(&s->gb); | |
4318 else | |
4319 sps->mb_aff= 0; | |
4320 | |
4321 sps->direct_8x8_inference_flag= get_bits1(&s->gb); | |
4322 | |
1371 | 4323 sps->crop= get_bits1(&s->gb); |
4324 if(sps->crop){ | |
4325 sps->crop_left = get_ue_golomb(&s->gb); | |
4326 sps->crop_right = get_ue_golomb(&s->gb); | |
4327 sps->crop_top = get_ue_golomb(&s->gb); | |
4328 sps->crop_bottom= get_ue_golomb(&s->gb); | |
4329 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
|
4330 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n"); |
1371 | 4331 } |
4332 }else{ | |
4333 sps->crop_left = | |
4334 sps->crop_right = | |
4335 sps->crop_top = | |
4336 sps->crop_bottom= 0; | |
4337 } | |
4338 | |
1168 | 4339 sps->vui_parameters_present_flag= get_bits1(&s->gb); |
4340 if( sps->vui_parameters_present_flag ) | |
4341 decode_vui_parameters(h, sps); | |
4342 | |
4343 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
|
4344 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 | 4345 sps_id, sps->profile_idc, sps->level_idc, |
4346 sps->poc_type, | |
4347 sps->ref_frame_count, | |
4348 sps->mb_width, sps->mb_height, | |
4349 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"), | |
4350 sps->direct_8x8_inference_flag ? "8B8" : "", | |
1371 | 4351 sps->crop_left, sps->crop_right, |
4352 sps->crop_top, sps->crop_bottom, | |
1168 | 4353 sps->vui_parameters_present_flag ? "VUI" : "" |
4354 ); | |
4355 } | |
4356 return 0; | |
4357 } | |
4358 | |
4359 static inline int decode_picture_parameter_set(H264Context *h){ | |
4360 MpegEncContext * const s = &h->s; | |
4361 int pps_id= get_ue_golomb(&s->gb); | |
4362 PPS *pps= &h->pps_buffer[pps_id]; | |
4363 | |
4364 pps->sps_id= get_ue_golomb(&s->gb); | |
4365 pps->cabac= get_bits1(&s->gb); | |
4366 pps->pic_order_present= get_bits1(&s->gb); | |
4367 pps->slice_group_count= get_ue_golomb(&s->gb) + 1; | |
4368 if(pps->slice_group_count > 1 ){ | |
4369 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
|
4370 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n"); |
1168 | 4371 switch(pps->mb_slice_group_map_type){ |
4372 case 0: | |
4373 #if 0 | |
4374 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | | | |
4375 | run_length[ i ] |1 |ue(v) | | |
4376 #endif | |
4377 break; | |
4378 case 2: | |
4379 #if 0 | |
4380 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | | | |
4381 |{ | | | | |
4382 | top_left_mb[ i ] |1 |ue(v) | | |
4383 | bottom_right_mb[ i ] |1 |ue(v) | | |
4384 | } | | | | |
4385 #endif | |
4386 break; | |
4387 case 3: | |
4388 case 4: | |
4389 case 5: | |
4390 #if 0 | |
4391 | slice_group_change_direction_flag |1 |u(1) | | |
4392 | slice_group_change_rate_minus1 |1 |ue(v) | | |
4393 #endif | |
4394 break; | |
4395 case 6: | |
4396 #if 0 | |
4397 | slice_group_id_cnt_minus1 |1 |ue(v) | | |
4398 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | | | |
4399 |) | | | | |
4400 | slice_group_id[ i ] |1 |u(v) | | |
4401 #endif | |
1214 | 4402 break; |
1168 | 4403 } |
4404 } | |
4405 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1; | |
4406 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1; | |
4407 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
|
4408 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n"); |
1168 | 4409 return -1; |
4410 } | |
4411 | |
4412 pps->weighted_pred= get_bits1(&s->gb); | |
4413 pps->weighted_bipred_idc= get_bits(&s->gb, 2); | |
4414 pps->init_qp= get_se_golomb(&s->gb) + 26; | |
4415 pps->init_qs= get_se_golomb(&s->gb) + 26; | |
4416 pps->chroma_qp_index_offset= get_se_golomb(&s->gb); | |
4417 pps->deblocking_filter_parameters_present= get_bits1(&s->gb); | |
4418 pps->constrained_intra_pred= get_bits1(&s->gb); | |
4419 pps->redundant_pic_cnt_present = get_bits1(&s->gb); | |
4420 | |
4421 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
|
4422 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 | 4423 pps_id, pps->sps_id, |
4424 pps->cabac ? "CABAC" : "CAVLC", | |
4425 pps->slice_group_count, | |
4426 pps->ref_count[0], pps->ref_count[1], | |
4427 pps->weighted_pred ? "weighted" : "", | |
4428 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset, | |
4429 pps->deblocking_filter_parameters_present ? "LPAR" : "", | |
4430 pps->constrained_intra_pred ? "CONSTR" : "", | |
1371 | 4431 pps->redundant_pic_cnt_present ? "REDU" : "" |
1168 | 4432 ); |
4433 } | |
4434 | |
4435 return 0; | |
4436 } | |
4437 | |
4438 /** | |
4439 * finds the end of the current frame in the bitstream. | |
4440 * @return the position of the first byte of the next frame, or -1 | |
4441 */ | |
4442 static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){ | |
4443 ParseContext *pc= &s->parse_context; | |
1187 | 4444 int i; |
1168 | 4445 uint32_t state; |
4446 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]); | |
4447 // mb_addr= pc->mb_addr - 1; | |
4448 state= pc->state; | |
4449 //FIXME this will fail with slices | |
4450 for(i=0; i<buf_size; i++){ | |
4451 state= (state<<8) | buf[i]; | |
4452 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){ | |
4453 if(pc->frame_start_found){ | |
4454 pc->state=-1; | |
4455 pc->frame_start_found= 0; | |
4456 return i-3; | |
4457 } | |
4458 pc->frame_start_found= 1; | |
4459 } | |
4460 } | |
4461 | |
4462 pc->state= state; | |
1219 | 4463 return END_NOT_FOUND; |
1168 | 4464 } |
4465 | |
4466 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){ | |
4467 MpegEncContext * const s = &h->s; | |
4468 AVCodecContext * const avctx= s->avctx; | |
4469 int buf_index=0; | |
1322 | 4470 #if 0 |
1168 | 4471 int i; |
4472 for(i=0; i<32; i++){ | |
4473 printf("%X ", buf[i]); | |
4474 } | |
4475 #endif | |
4476 for(;;){ | |
4477 int consumed; | |
4478 int dst_length; | |
4479 int bit_length; | |
4480 uint8_t *ptr; | |
4481 | |
4482 // start code prefix search | |
4483 for(; buf_index + 3 < buf_size; buf_index++){ | |
4484 // this should allways succeed in the first iteration | |
4485 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1) | |
4486 break; | |
4487 } | |
4488 | |
4489 if(buf_index+3 >= buf_size) break; | |
4490 | |
4491 buf_index+=3; | |
4492 | |
4493 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index); | |
4494 if(ptr[dst_length - 1] == 0) dst_length--; | |
4495 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1); | |
4496 | |
4497 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
|
4498 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length); |
1168 | 4499 } |
4500 | |
4501 buf_index += consumed; | |
4502 | |
4503 if(h->nal_ref_idc < s->hurry_up) | |
4504 continue; | |
4505 | |
4506 switch(h->nal_unit_type){ | |
4507 case NAL_IDR_SLICE: | |
4508 idr(h); //FIXME ensure we dont loose some frames if there is reordering | |
4509 case NAL_SLICE: | |
4510 init_get_bits(&s->gb, ptr, bit_length); | |
4511 h->intra_gb_ptr= | |
4512 h->inter_gb_ptr= &s->gb; | |
4513 s->data_partitioning = 0; | |
4514 | |
4515 if(decode_slice_header(h) < 0) return -1; | |
4516 if(h->redundant_pic_count==0) | |
4517 decode_slice(h); | |
4518 break; | |
4519 case NAL_DPA: | |
4520 init_get_bits(&s->gb, ptr, bit_length); | |
4521 h->intra_gb_ptr= | |
4522 h->inter_gb_ptr= NULL; | |
4523 s->data_partitioning = 1; | |
4524 | |
4525 if(decode_slice_header(h) < 0) return -1; | |
4526 break; | |
4527 case NAL_DPB: | |
4528 init_get_bits(&h->intra_gb, ptr, bit_length); | |
4529 h->intra_gb_ptr= &h->intra_gb; | |
4530 break; | |
4531 case NAL_DPC: | |
4532 init_get_bits(&h->inter_gb, ptr, bit_length); | |
4533 h->inter_gb_ptr= &h->inter_gb; | |
1174 | 4534 |
4535 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning) | |
1168 | 4536 decode_slice(h); |
4537 break; | |
4538 case NAL_SEI: | |
4539 break; | |
4540 case NAL_SPS: | |
4541 init_get_bits(&s->gb, ptr, bit_length); | |
4542 decode_seq_parameter_set(h); | |
4543 | |
4544 if(s->flags& CODEC_FLAG_LOW_DELAY) | |
4545 s->low_delay=1; | |
4546 | |
4547 avctx->has_b_frames= !s->low_delay; | |
4548 break; | |
4549 case NAL_PPS: | |
4550 init_get_bits(&s->gb, ptr, bit_length); | |
4551 | |
4552 decode_picture_parameter_set(h); | |
4553 | |
4554 break; | |
4555 case NAL_PICTURE_DELIMITER: | |
4556 break; | |
4557 case NAL_FILTER_DATA: | |
4558 break; | |
4559 } | |
4560 | |
4561 //FIXME move after where irt is set | |
4562 s->current_picture.pict_type= s->pict_type; | |
4563 s->current_picture.key_frame= s->pict_type == I_TYPE; | |
4564 } | |
4565 | |
1174 | 4566 if(!s->current_picture_ptr) return buf_index; //no frame |
4567 | |
1168 | 4568 h->prev_frame_num_offset= h->frame_num_offset; |
4569 h->prev_frame_num= h->frame_num; | |
4570 if(s->current_picture_ptr->reference){ | |
4571 h->prev_poc_msb= h->poc_msb; | |
4572 h->prev_poc_lsb= h->poc_lsb; | |
4573 } | |
4574 if(s->current_picture_ptr->reference) | |
4575 execute_ref_pic_marking(h, h->mmco, h->mmco_index); | |
4576 else | |
1169 | 4577 assert(h->mmco_index==0); |
1168 | 4578 |
4579 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
|
4580 |
1168 | 4581 MPV_frame_end(s); |
4582 | |
4583 return buf_index; | |
4584 } | |
4585 | |
4586 /** | |
4587 * retunrs the number of bytes consumed for building the current frame | |
4588 */ | |
4589 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){ | |
4590 if(s->flags&CODEC_FLAG_TRUNCATED){ | |
4591 pos -= s->parse_context.last_index; | |
4592 if(pos<0) pos=0; // FIXME remove (uneeded?) | |
4593 | |
4594 return pos; | |
4595 }else{ | |
4596 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...) | |
4597 if(pos+10>buf_size) pos=buf_size; // oops ;) | |
4598 | |
4599 return pos; | |
4600 } | |
4601 } | |
4602 | |
4603 static int decode_frame(AVCodecContext *avctx, | |
4604 void *data, int *data_size, | |
4605 uint8_t *buf, int buf_size) | |
4606 { | |
4607 H264Context *h = avctx->priv_data; | |
4608 MpegEncContext *s = &h->s; | |
4609 AVFrame *pict = data; | |
4610 int buf_index; | |
4611 | |
4612 s->flags= avctx->flags; | |
1754
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1706
diff
changeset
|
4613 s->flags2= avctx->flags2; |
1168 | 4614 |
4615 *data_size = 0; | |
4616 | |
4617 /* no supplementary picture */ | |
4618 if (buf_size == 0) { | |
4619 return 0; | |
4620 } | |
4621 | |
4622 if(s->flags&CODEC_FLAG_TRUNCATED){ | |
4623 int next= find_frame_end(s, buf, buf_size); | |
4624 | |
4625 if( ff_combine_frame(s, next, &buf, &buf_size) < 0 ) | |
4626 return buf_size; | |
4627 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index); | |
4628 } | |
4629 | |
4630 if(s->avctx->extradata_size && s->picture_number==0){ | |
4631 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) ) | |
4632 return -1; | |
4633 } | |
4634 | |
4635 buf_index=decode_nal_units(h, buf, buf_size); | |
4636 if(buf_index < 0) | |
4637 return -1; | |
4638 | |
4639 //FIXME do something with unavailable reference frames | |
4640 | |
4641 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size); | |
4642 #if 0 | |
4643 if(s->pict_type==B_TYPE || s->low_delay){ | |
4644 *pict= *(AVFrame*)&s->current_picture; | |
4645 } else { | |
4646 *pict= *(AVFrame*)&s->last_picture; | |
4647 } | |
4648 #endif | |
1174 | 4649 if(!s->current_picture_ptr){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4650 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n"); |
1174 | 4651 return -1; |
4652 } | |
4653 | |
1168 | 4654 *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
|
4655 ff_print_debug_info(s, pict); |
1168 | 4656 assert(pict->data[0]); |
4657 //printf("out %d\n", (int)pict->data[0]); | |
4658 #if 0 //? | |
4659 | |
4660 /* Return the Picture timestamp as the frame number */ | |
4661 /* we substract 1 because it is added on utils.c */ | |
4662 avctx->frame_number = s->picture_number - 1; | |
4663 #endif | |
4664 #if 0 | |
4665 /* dont output the last pic after seeking */ | |
4666 if(s->last_picture_ptr || s->low_delay) | |
4667 //Note this isnt a issue as a IDR pic should flush teh buffers | |
4668 #endif | |
4669 *data_size = sizeof(AVFrame); | |
4670 return get_consumed_bytes(s, buf_index, buf_size); | |
4671 } | |
4672 #if 0 | |
4673 static inline void fill_mb_avail(H264Context *h){ | |
4674 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
|
4675 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 4676 |
4677 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
|
4678 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
|
4679 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
|
4680 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num; |
1168 | 4681 }else{ |
4682 h->mb_avail[0]= | |
4683 h->mb_avail[1]= | |
4684 h->mb_avail[2]= 0; | |
4685 } | |
4686 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num; | |
4687 h->mb_avail[4]= 1; //FIXME move out | |
4688 h->mb_avail[5]= 0; //FIXME move out | |
4689 } | |
4690 #endif | |
4691 | |
4692 #if 0 //selftest | |
4693 #define COUNT 8000 | |
4694 #define SIZE (COUNT*40) | |
4695 int main(){ | |
4696 int i; | |
4697 uint8_t temp[SIZE]; | |
4698 PutBitContext pb; | |
4699 GetBitContext gb; | |
4700 // int int_temp[10000]; | |
4701 DSPContext dsp; | |
4702 AVCodecContext avctx; | |
4703 | |
4704 dsputil_init(&dsp, &avctx); | |
4705 | |
1522
79dddc5cd990
removed the obsolete and unused parameters of init_put_bits
alex
parents:
1453
diff
changeset
|
4706 init_put_bits(&pb, temp, SIZE); |
1168 | 4707 printf("testing unsigned exp golomb\n"); |
4708 for(i=0; i<COUNT; i++){ | |
4709 START_TIMER | |
4710 set_ue_golomb(&pb, i); | |
4711 STOP_TIMER("set_ue_golomb"); | |
4712 } | |
4713 flush_put_bits(&pb); | |
4714 | |
4715 init_get_bits(&gb, temp, 8*SIZE); | |
4716 for(i=0; i<COUNT; i++){ | |
4717 int j, s; | |
4718 | |
4719 s= show_bits(&gb, 24); | |
4720 | |
4721 START_TIMER | |
4722 j= get_ue_golomb(&gb); | |
4723 if(j != i){ | |
4724 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s); | |
4725 // return -1; | |
4726 } | |
4727 STOP_TIMER("get_ue_golomb"); | |
4728 } | |
4729 | |
4730 | |
1524 | 4731 init_put_bits(&pb, temp, SIZE); |
1168 | 4732 printf("testing signed exp golomb\n"); |
4733 for(i=0; i<COUNT; i++){ | |
4734 START_TIMER | |
4735 set_se_golomb(&pb, i - COUNT/2); | |
4736 STOP_TIMER("set_se_golomb"); | |
4737 } | |
4738 flush_put_bits(&pb); | |
4739 | |
4740 init_get_bits(&gb, temp, 8*SIZE); | |
4741 for(i=0; i<COUNT; i++){ | |
4742 int j, s; | |
4743 | |
4744 s= show_bits(&gb, 24); | |
4745 | |
4746 START_TIMER | |
4747 j= get_se_golomb(&gb); | |
4748 if(j != i - COUNT/2){ | |
4749 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s); | |
4750 // return -1; | |
4751 } | |
4752 STOP_TIMER("get_se_golomb"); | |
4753 } | |
4754 | |
4755 printf("testing 4x4 (I)DCT\n"); | |
4756 | |
4757 DCTELEM block[16]; | |
4758 uint8_t src[16], ref[16]; | |
4759 uint64_t error= 0, max_error=0; | |
4760 | |
4761 for(i=0; i<COUNT; i++){ | |
4762 int j; | |
4763 // printf("%d %d %d\n", r1, r2, (r2-r1)*16); | |
4764 for(j=0; j<16; j++){ | |
4765 ref[j]= random()%255; | |
4766 src[j]= random()%255; | |
4767 } | |
4768 | |
4769 h264_diff_dct_c(block, src, ref, 4); | |
4770 | |
4771 //normalize | |
4772 for(j=0; j<16; j++){ | |
4773 // printf("%d ", block[j]); | |
4774 block[j]= block[j]*4; | |
4775 if(j&1) block[j]= (block[j]*4 + 2)/5; | |
4776 if(j&4) block[j]= (block[j]*4 + 2)/5; | |
4777 } | |
4778 // printf("\n"); | |
4779 | |
4780 h264_add_idct_c(ref, block, 4); | |
4781 /* for(j=0; j<16; j++){ | |
4782 printf("%d ", ref[j]); | |
4783 } | |
4784 printf("\n");*/ | |
4785 | |
4786 for(j=0; j<16; j++){ | |
4787 int diff= ABS(src[j] - ref[j]); | |
4788 | |
4789 error+= diff*diff; | |
4790 max_error= FFMAX(max_error, diff); | |
4791 } | |
4792 } | |
4793 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error ); | |
4794 #if 0 | |
4795 printf("testing quantizer\n"); | |
4796 for(qp=0; qp<52; qp++){ | |
4797 for(i=0; i<16; i++) | |
4798 src1_block[i]= src2_block[i]= random()%255; | |
4799 | |
4800 } | |
4801 #endif | |
4802 printf("Testing NAL layer\n"); | |
4803 | |
4804 uint8_t bitstream[COUNT]; | |
4805 uint8_t nal[COUNT*2]; | |
4806 H264Context h; | |
4807 memset(&h, 0, sizeof(H264Context)); | |
4808 | |
4809 for(i=0; i<COUNT; i++){ | |
4810 int zeros= i; | |
4811 int nal_length; | |
4812 int consumed; | |
4813 int out_length; | |
4814 uint8_t *out; | |
4815 int j; | |
4816 | |
4817 for(j=0; j<COUNT; j++){ | |
4818 bitstream[j]= (random() % 255) + 1; | |
4819 } | |
4820 | |
4821 for(j=0; j<zeros; j++){ | |
4822 int pos= random() % COUNT; | |
4823 while(bitstream[pos] == 0){ | |
4824 pos++; | |
4825 pos %= COUNT; | |
4826 } | |
4827 bitstream[pos]=0; | |
4828 } | |
4829 | |
4830 START_TIMER | |
4831 | |
4832 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2); | |
4833 if(nal_length<0){ | |
4834 printf("encoding failed\n"); | |
4835 return -1; | |
4836 } | |
4837 | |
4838 out= decode_nal(&h, nal, &out_length, &consumed, nal_length); | |
4839 | |
4840 STOP_TIMER("NAL") | |
4841 | |
4842 if(out_length != COUNT){ | |
4843 printf("incorrect length %d %d\n", out_length, COUNT); | |
4844 return -1; | |
4845 } | |
4846 | |
4847 if(consumed != nal_length){ | |
4848 printf("incorrect consumed length %d %d\n", nal_length, consumed); | |
4849 return -1; | |
4850 } | |
4851 | |
4852 if(memcmp(bitstream, out, COUNT)){ | |
4853 printf("missmatch\n"); | |
4854 return -1; | |
4855 } | |
4856 } | |
4857 | |
4858 printf("Testing RBSP\n"); | |
4859 | |
4860 | |
4861 return 0; | |
4862 } | |
4863 #endif | |
4864 | |
4865 | |
4866 static int decode_end(AVCodecContext *avctx) | |
4867 { | |
4868 H264Context *h = avctx->priv_data; | |
4869 MpegEncContext *s = &h->s; | |
4870 | |
4871 free_tables(h); //FIXME cleanup init stuff perhaps | |
4872 MPV_common_end(s); | |
4873 | |
4874 // memset(h, 0, sizeof(H264Context)); | |
4875 | |
4876 return 0; | |
4877 } | |
4878 | |
4879 | |
4880 AVCodec h264_decoder = { | |
4881 "h264", | |
4882 CODEC_TYPE_VIDEO, | |
4883 CODEC_ID_H264, | |
4884 sizeof(H264Context), | |
4885 decode_init, | |
4886 NULL, | |
4887 decode_end, | |
4888 decode_frame, | |
1214 | 4889 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED, |
1168 | 4890 }; |
4891 | |
1234 | 4892 #include "svq3.c" |