1168
|
1 /*
|
|
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
|
|
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
|
|
4 *
|
|
5 * This library is free software; you can redistribute it and/or
|
|
6 * modify it under the terms of the GNU Lesser General Public
|
|
7 * License as published by the Free Software Foundation; either
|
|
8 * version 2 of the License, or (at your option) any later version.
|
|
9 *
|
|
10 * This library is distributed in the hope that it will be useful,
|
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
13 * Lesser General Public License for more details.
|
|
14 *
|
|
15 * You should have received a copy of the GNU Lesser General Public
|
|
16 * License along with this library; if not, write to the Free Software
|
|
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
18 *
|
|
19 */
|
|
20
|
|
21 /**
|
|
22 * @file h264.c
|
|
23 * H.264 / AVC / MPEG4 part10 codec.
|
|
24 * @author Michael Niedermayer <michaelni@gmx.at>
|
|
25 */
|
|
26
|
|
27 #include "common.h"
|
|
28 #include "dsputil.h"
|
|
29 #include "avcodec.h"
|
|
30 #include "mpegvideo.h"
|
|
31 #include "h264data.h"
|
|
32 #include "golomb.h"
|
|
33
|
|
34 #undef NDEBUG
|
|
35 #include <assert.h>
|
|
36
|
|
37 #define interlaced_dct interlaced_dct_is_a_bad_name
|
|
38 #define mb_intra mb_intra_isnt_initalized_see_mb_type
|
|
39
|
|
40 #define LUMA_DC_BLOCK_INDEX 25
|
|
41 #define CHROMA_DC_BLOCK_INDEX 26
|
|
42
|
|
43 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
|
|
44 #define COEFF_TOKEN_VLC_BITS 8
|
|
45 #define TOTAL_ZEROS_VLC_BITS 9
|
|
46 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
|
|
47 #define RUN_VLC_BITS 3
|
|
48 #define RUN7_VLC_BITS 6
|
|
49
|
|
50 #define MAX_SPS_COUNT 32
|
|
51 #define MAX_PPS_COUNT 256
|
|
52
|
|
53 #define MAX_MMCO_COUNT 66
|
|
54
|
|
55 /**
|
|
56 * Sequence parameter set
|
|
57 */
|
|
58 typedef struct SPS{
|
|
59
|
|
60 int profile_idc;
|
|
61 int level_idc;
|
|
62 int multiple_slice_groups; ///< more_than_one_slice_group_allowed_flag
|
|
63 int arbitrary_slice_order; ///< arbitrary_slice_order_allowed_flag
|
|
64 int redundant_slices; ///< redundant_slices_allowed_flag
|
|
65 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
|
|
66 int poc_type; ///< pic_order_cnt_type
|
|
67 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
|
|
68 int delta_pic_order_always_zero_flag;
|
|
69 int offset_for_non_ref_pic;
|
|
70 int offset_for_top_to_bottom_field;
|
|
71 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
|
|
72 int ref_frame_count; ///< num_ref_frames
|
|
73 int required_frame_num_update_behaviour_flag;
|
|
74 int mb_width; ///< frame_width_in_mbs_minus1 + 1
|
|
75 int mb_height; ///< frame_height_in_mbs_minus1 + 1
|
|
76 int frame_mbs_only_flag;
|
|
77 int mb_aff; ///<mb_adaptive_frame_field_flag
|
|
78 int direct_8x8_inference_flag;
|
|
79 int vui_parameters_present_flag;
|
|
80 int sar_width;
|
|
81 int sar_height;
|
|
82 short offset_for_ref_frame[256]; //FIXME dyn aloc?
|
|
83 }SPS;
|
|
84
|
|
85 /**
|
|
86 * Picture parameter set
|
|
87 */
|
|
88 typedef struct PPS{
|
|
89 int sps_id;
|
|
90 int cabac; ///< entropy_coding_mode_flag
|
|
91 int pic_order_present; ///< pic_order_present_flag
|
|
92 int slice_group_count; ///< num_slice_groups_minus1 + 1
|
|
93 int mb_slice_group_map_type;
|
|
94 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
|
|
95 int weighted_pred; ///< weighted_pred_flag
|
|
96 int weighted_bipred_idc;
|
|
97 int init_qp; ///< pic_init_qp_minus26 + 26
|
|
98 int init_qs; ///< pic_init_qs_minus26 + 26
|
|
99 int chroma_qp_index_offset;
|
|
100 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
|
|
101 int constrained_intra_pred; ///< constrained_intra_pred_flag
|
|
102 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
|
|
103 int crop; ///< frame_cropping_flag
|
|
104 int crop_left; ///< frame_cropping_rect_left_offset
|
|
105 int crop_right; ///< frame_cropping_rect_right_offset
|
|
106 int crop_top; ///< frame_cropping_rect_top_offset
|
|
107 int crop_bottom; ///< frame_cropping_rect_bottom_offset
|
|
108 }PPS;
|
|
109
|
|
110 /**
|
|
111 * Memory management control operation opcode.
|
|
112 */
|
|
113 typedef enum MMCOOpcode{
|
|
114 MMCO_END=0,
|
|
115 MMCO_SHORT2UNUSED,
|
|
116 MMCO_LONG2UNUSED,
|
|
117 MMCO_SHORT2LONG,
|
|
118 MMCO_SET_MAX_LONG,
|
|
119 MMCO_RESET,
|
|
120 MMCO_LONG,
|
|
121 } MMCOOpcode;
|
|
122
|
|
123 /**
|
|
124 * Memory management control operation.
|
|
125 */
|
|
126 typedef struct MMCO{
|
|
127 MMCOOpcode opcode;
|
|
128 int short_frame_num;
|
|
129 int long_index;
|
|
130 } MMCO;
|
|
131
|
|
132 /**
|
|
133 * H264Context
|
|
134 */
|
|
135 typedef struct H264Context{
|
|
136 MpegEncContext s;
|
|
137 int nal_ref_idc;
|
|
138 int nal_unit_type;
|
|
139 #define NAL_SLICE 1
|
|
140 #define NAL_DPA 2
|
|
141 #define NAL_DPB 3
|
|
142 #define NAL_DPC 4
|
|
143 #define NAL_IDR_SLICE 5
|
|
144 #define NAL_SEI 6
|
|
145 #define NAL_SPS 7
|
|
146 #define NAL_PPS 8
|
|
147 #define NAL_PICTURE_DELIMITER 9
|
|
148 #define NAL_FILTER_DATA 10
|
|
149 uint8_t *rbsp_buffer;
|
|
150 int rbsp_buffer_size;
|
|
151
|
|
152 int mb_stride; ///< stride of some mb tables
|
|
153
|
|
154 int chroma_qp; //QPc
|
|
155
|
|
156 int prev_mb_skiped; //FIXME remove (IMHO not used)
|
|
157
|
|
158 //prediction stuff
|
|
159 int chroma_pred_mode;
|
|
160 int intra16x16_pred_mode;
|
|
161
|
|
162 int8_t intra4x4_pred_mode_cache[5*8];
|
|
163 int8_t (*intra4x4_pred_mode)[8];
|
|
164 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
|
|
165 void (*pred8x8 [4+3])(uint8_t *src, int stride);
|
|
166 void (*pred16x16[4+3])(uint8_t *src, int stride);
|
|
167 unsigned int topleft_samples_available;
|
|
168 unsigned int top_samples_available;
|
|
169 unsigned int topright_samples_available;
|
|
170 unsigned int left_samples_available;
|
|
171
|
|
172 /**
|
|
173 * non zero coeff count cache.
|
|
174 * is 64 if not available.
|
|
175 */
|
|
176 uint8_t non_zero_count_cache[6*8];
|
|
177 uint8_t (*non_zero_count)[16];
|
|
178
|
|
179 /**
|
|
180 * Motion vector cache.
|
|
181 */
|
|
182 int16_t mv_cache[2][5*8][2];
|
|
183 int8_t ref_cache[2][5*8];
|
|
184 #define LIST_NOT_USED -1 //FIXME rename?
|
|
185 #define PART_NOT_AVAILABLE -2
|
|
186
|
|
187 /**
|
|
188 * is 1 if the specific list MV&references are set to 0,0,-2.
|
|
189 */
|
|
190 int mv_cache_clean[2];
|
|
191
|
|
192 int block_offset[16+8];
|
|
193 int chroma_subblock_offset[16]; //FIXME remove
|
|
194
|
|
195 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
|
|
196 uint16_t *mb2b8_xy;
|
|
197 int b_stride;
|
|
198 int b8_stride;
|
|
199
|
|
200 SPS sps_buffer[MAX_SPS_COUNT];
|
|
201 SPS sps; ///< current sps
|
|
202
|
|
203 PPS pps_buffer[MAX_PPS_COUNT];
|
|
204 /**
|
|
205 * current pps
|
|
206 */
|
|
207 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
|
|
208
|
|
209 int slice_num;
|
|
210 uint8_t *slice_table_base;
|
|
211 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
|
|
212 int slice_type;
|
|
213 int slice_type_fixed;
|
|
214
|
|
215 //interlacing specific flags
|
|
216 int mb_field_decoding_flag;
|
|
217
|
|
218 int sub_mb_type[4];
|
|
219
|
|
220 //POC stuff
|
|
221 int poc_lsb;
|
|
222 int poc_msb;
|
|
223 int delta_poc_bottom;
|
|
224 int delta_poc[2];
|
|
225 int frame_num;
|
|
226 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
|
|
227 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
|
|
228 int frame_num_offset; ///< for POC type 2
|
|
229 int prev_frame_num_offset; ///< for POC type 2
|
|
230 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
|
|
231
|
|
232 /**
|
|
233 * frame_num for frames or 2*frame_num for field pics.
|
|
234 */
|
|
235 int curr_pic_num;
|
|
236
|
|
237 /**
|
|
238 * max_frame_num or 2*max_frame_num for field pics.
|
|
239 */
|
|
240 int max_pic_num;
|
|
241
|
|
242 //Weighted pred stuff
|
|
243 int luma_log2_weight_denom;
|
|
244 int chroma_log2_weight_denom;
|
|
245 int luma_weight[2][16];
|
|
246 int luma_offset[2][16];
|
|
247 int chroma_weight[2][16][2];
|
|
248 int chroma_offset[2][16][2];
|
|
249
|
|
250 //deblock
|
|
251 int disable_deblocking_filter_idc;
|
|
252 int slice_alpha_c0_offset_div2;
|
|
253 int slice_beta_offset_div2;
|
|
254
|
|
255 int redundant_pic_count;
|
|
256
|
|
257 int direct_spatial_mv_pred;
|
|
258
|
|
259 /**
|
|
260 * num_ref_idx_l0/1_active_minus1 + 1
|
|
261 */
|
|
262 int ref_count[2];// FIXME split for AFF
|
|
263 Picture *short_ref[16];
|
|
264 Picture *long_ref[16];
|
|
265 Picture default_ref_list[2][32];
|
|
266 Picture ref_list[2][32]; //FIXME size?
|
|
267 Picture field_ref_list[2][32]; //FIXME size?
|
|
268
|
|
269 /**
|
|
270 * memory management control operations buffer.
|
|
271 */
|
|
272 MMCO mmco[MAX_MMCO_COUNT];
|
|
273 int mmco_index;
|
|
274
|
|
275 int long_ref_count; ///< number of actual long term references
|
|
276 int short_ref_count; ///< number of actual short term references
|
|
277
|
|
278 //data partitioning
|
|
279 GetBitContext intra_gb;
|
|
280 GetBitContext inter_gb;
|
|
281 GetBitContext *intra_gb_ptr;
|
|
282 GetBitContext *inter_gb_ptr;
|
|
283
|
|
284 DCTELEM mb[16*24] __align8;
|
|
285 }H264Context;
|
|
286
|
|
287 static VLC coeff_token_vlc[4];
|
|
288 static VLC chroma_dc_coeff_token_vlc;
|
|
289
|
|
290 static VLC total_zeros_vlc[15];
|
|
291 static VLC chroma_dc_total_zeros_vlc[3];
|
|
292
|
|
293 static VLC run_vlc[6];
|
|
294 static VLC run7_vlc;
|
|
295
|
|
296 /**
|
|
297 * fill a rectangle.
|
|
298 * @param h height of the recatangle, should be a constant
|
|
299 * @param w width of the recatangle, should be a constant
|
|
300 * @param size the size of val (1 or 4), should be a constant
|
|
301 */
|
|
302 static inline void fill_rectangle(void *p, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
|
|
303 assert(size==1 || size==4);
|
|
304
|
|
305 w *= size;
|
|
306 stride *= size;
|
|
307
|
|
308 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
|
|
309 if(w==2 && h==2){
|
|
310 *(uint16_t*)(p + 0)=
|
|
311 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
|
|
312 }else if(w==2 && h==4){
|
|
313 *(uint16_t*)(p + 0*stride)=
|
|
314 *(uint16_t*)(p + 1*stride)=
|
|
315 *(uint16_t*)(p + 2*stride)=
|
|
316 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
|
|
317 }else if(w==4 && h==2){
|
|
318 *(uint32_t*)(p + 0*stride)=
|
|
319 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
|
|
320 }else if(w==4 && h==4){
|
|
321 *(uint32_t*)(p + 0*stride)=
|
|
322 *(uint32_t*)(p + 1*stride)=
|
|
323 *(uint32_t*)(p + 2*stride)=
|
|
324 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
|
|
325 }else if(w==8 && h==1){
|
|
326 *(uint32_t*)(p + 0)=
|
|
327 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
|
|
328 }else if(w==8 && h==2){
|
|
329 *(uint32_t*)(p + 0 + 0*stride)=
|
|
330 *(uint32_t*)(p + 4 + 0*stride)=
|
|
331 *(uint32_t*)(p + 0 + 1*stride)=
|
|
332 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
|
|
333 }else if(w==8 && h==4){
|
|
334 *(uint64_t*)(p + 0*stride)=
|
|
335 *(uint64_t*)(p + 1*stride)=
|
|
336 *(uint64_t*)(p + 2*stride)=
|
|
337 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
|
|
338 }else if(w==16 && h==2){
|
|
339 *(uint64_t*)(p + 0+0*stride)=
|
|
340 *(uint64_t*)(p + 8+0*stride)=
|
|
341 *(uint64_t*)(p + 0+1*stride)=
|
|
342 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
|
|
343 }else if(w==16 && h==4){
|
|
344 *(uint64_t*)(p + 0+0*stride)=
|
|
345 *(uint64_t*)(p + 8+0*stride)=
|
|
346 *(uint64_t*)(p + 0+1*stride)=
|
|
347 *(uint64_t*)(p + 8+1*stride)=
|
|
348 *(uint64_t*)(p + 0+2*stride)=
|
|
349 *(uint64_t*)(p + 8+2*stride)=
|
|
350 *(uint64_t*)(p + 0+3*stride)=
|
|
351 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
|
|
352 }else
|
|
353 assert(0);
|
|
354 }
|
|
355
|
|
356 static inline void fill_caches(H264Context *h, int mb_type){
|
|
357 MpegEncContext * const s = &h->s;
|
|
358 const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
|
|
359 int topleft_xy, top_xy, topright_xy, left_xy[2];
|
|
360 int topleft_type, top_type, topright_type, left_type[2];
|
|
361 int left_block[4];
|
|
362 int i;
|
|
363
|
|
364 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
|
|
365
|
|
366 if(h->sps.mb_aff){
|
|
367 //FIXME
|
|
368 }else{
|
|
369 topleft_xy = mb_xy-1 - h->mb_stride;
|
|
370 top_xy = mb_xy - h->mb_stride;
|
|
371 topright_xy= mb_xy+1 - h->mb_stride;
|
|
372 left_xy[0] = mb_xy-1;
|
|
373 left_xy[1] = mb_xy-1;
|
|
374 left_block[0]= 0;
|
|
375 left_block[1]= 1;
|
|
376 left_block[2]= 2;
|
|
377 left_block[3]= 3;
|
|
378 }
|
|
379
|
|
380 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
|
|
381 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
|
|
382 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
|
|
383 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
|
|
384 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
|
|
385
|
|
386 if(IS_INTRA(mb_type)){
|
|
387 h->topleft_samples_available=
|
|
388 h->top_samples_available=
|
|
389 h->left_samples_available= 0xFFFF;
|
|
390 h->topright_samples_available= 0xEEEA;
|
|
391
|
|
392 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
|
|
393 h->topleft_samples_available= 0xB3FF;
|
|
394 h->top_samples_available= 0x33FF;
|
|
395 h->topright_samples_available= 0x26EA;
|
|
396 }
|
|
397 for(i=0; i<2; i++){
|
|
398 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
|
|
399 h->topleft_samples_available&= 0xDF5F;
|
|
400 h->left_samples_available&= 0x5F5F;
|
|
401 }
|
|
402 }
|
|
403
|
|
404 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
|
|
405 h->topleft_samples_available&= 0x7FFF;
|
|
406
|
|
407 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
|
|
408 h->topright_samples_available&= 0xFBFF;
|
|
409
|
|
410 if(IS_INTRA4x4(mb_type)){
|
|
411 if(IS_INTRA4x4(top_type)){
|
|
412 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
|
|
413 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
|
|
414 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
|
|
415 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
|
|
416 }else{
|
|
417 int pred;
|
|
418 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
|
|
419 pred= 2;
|
|
420 else{
|
|
421 pred= -1;
|
|
422 }
|
|
423 h->intra4x4_pred_mode_cache[4+8*0]=
|
|
424 h->intra4x4_pred_mode_cache[5+8*0]=
|
|
425 h->intra4x4_pred_mode_cache[6+8*0]=
|
|
426 h->intra4x4_pred_mode_cache[7+8*0]= pred;
|
|
427 }
|
|
428 for(i=0; i<2; i++){
|
|
429 if(IS_INTRA4x4(left_type[i])){
|
|
430 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
|
|
431 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
|
|
432 }else{
|
|
433 int pred;
|
|
434 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
|
|
435 pred= 2;
|
|
436 else{
|
|
437 pred= -1;
|
|
438 }
|
|
439 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
|
|
440 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
|
|
441 }
|
|
442 }
|
|
443 }
|
|
444 }
|
|
445
|
|
446
|
|
447 /*
|
|
448 0 . T T. T T T T
|
|
449 1 L . .L . . . .
|
|
450 2 L . .L . . . .
|
|
451 3 . T TL . . . .
|
|
452 4 L . .L . . . .
|
|
453 5 L . .. . . . .
|
|
454 */
|
|
455 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
|
|
456 if(top_type){
|
|
457 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
|
|
458 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
|
|
459 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
|
|
460 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
|
|
461
|
|
462 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
|
|
463 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
|
|
464
|
|
465 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
|
|
466 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
|
|
467 }else{
|
|
468 h->non_zero_count_cache[4+8*0]=
|
|
469 h->non_zero_count_cache[5+8*0]=
|
|
470 h->non_zero_count_cache[6+8*0]=
|
|
471 h->non_zero_count_cache[7+8*0]=
|
|
472
|
|
473 h->non_zero_count_cache[1+8*0]=
|
|
474 h->non_zero_count_cache[2+8*0]=
|
|
475
|
|
476 h->non_zero_count_cache[1+8*3]=
|
|
477 h->non_zero_count_cache[2+8*3]= 64;
|
|
478 }
|
|
479
|
|
480 if(left_type[0]){
|
|
481 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
|
|
482 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
|
|
483 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
|
|
484 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
|
|
485 }else{
|
|
486 h->non_zero_count_cache[3+8*1]=
|
|
487 h->non_zero_count_cache[3+8*2]=
|
|
488 h->non_zero_count_cache[0+8*1]=
|
|
489 h->non_zero_count_cache[0+8*4]= 64;
|
|
490 }
|
|
491
|
|
492 if(left_type[1]){
|
|
493 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
|
|
494 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
|
|
495 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
|
|
496 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
|
|
497 }else{
|
|
498 h->non_zero_count_cache[3+8*3]=
|
|
499 h->non_zero_count_cache[3+8*4]=
|
|
500 h->non_zero_count_cache[0+8*2]=
|
|
501 h->non_zero_count_cache[0+8*5]= 64;
|
|
502 }
|
|
503
|
|
504 #if 1
|
|
505 if(IS_INTER(mb_type)){
|
|
506 int list;
|
|
507 for(list=0; list<2; list++){
|
|
508 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
|
|
509 /*if(!h->mv_cache_clean[list]){
|
|
510 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
|
|
511 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
|
|
512 h->mv_cache_clean[list]= 1;
|
|
513 }*/
|
|
514 continue; //FIXME direct mode ...
|
|
515 }
|
|
516 h->mv_cache_clean[list]= 0;
|
|
517
|
|
518 if(IS_INTER(topleft_type)){
|
|
519 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
|
|
520 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
|
|
521 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
|
|
522 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
|
|
523 }else{
|
|
524 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
|
|
525 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
|
|
526 }
|
|
527
|
|
528 if(IS_INTER(top_type)){
|
|
529 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
|
|
530 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
|
|
531 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
|
|
532 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
|
|
533 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
|
|
534 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
|
|
535 h->ref_cache[list][scan8[0] + 0 - 1*8]=
|
|
536 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
|
|
537 h->ref_cache[list][scan8[0] + 2 - 1*8]=
|
|
538 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
|
|
539 }else{
|
|
540 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
|
|
541 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
|
|
542 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
|
|
543 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
|
|
544 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
|
|
545 }
|
|
546
|
|
547 if(IS_INTER(topright_type)){
|
|
548 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
|
|
549 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
|
|
550 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
|
|
551 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
|
|
552 }else{
|
|
553 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
|
|
554 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
|
|
555 }
|
|
556
|
|
557 //FIXME unify cleanup or sth
|
|
558 if(IS_INTER(left_type[0])){
|
|
559 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
|
|
560 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
|
|
561 *(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]];
|
|
562 *(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]];
|
|
563 h->ref_cache[list][scan8[0] - 1 + 0*8]=
|
|
564 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
|
|
565 }else{
|
|
566 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
|
|
567 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
|
|
568 h->ref_cache[list][scan8[0] - 1 + 0*8]=
|
|
569 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
|
|
570 }
|
|
571
|
|
572 if(IS_INTER(left_type[1])){
|
|
573 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
|
|
574 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
|
|
575 *(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]];
|
|
576 *(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]];
|
|
577 h->ref_cache[list][scan8[0] - 1 + 2*8]=
|
|
578 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
|
|
579 }else{
|
|
580 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
|
|
581 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
|
|
582 h->ref_cache[list][scan8[0] - 1 + 2*8]=
|
|
583 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
|
|
584 }
|
|
585
|
|
586 h->ref_cache[list][scan8[5 ]+1] =
|
|
587 h->ref_cache[list][scan8[7 ]+1] =
|
|
588 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
|
|
589 h->ref_cache[list][scan8[4 ]] =
|
|
590 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
|
|
591 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
|
|
592 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
|
|
593 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
|
|
594 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
|
|
595 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
|
|
596 }
|
|
597 //FIXME
|
|
598
|
|
599 }
|
|
600 #endif
|
|
601 }
|
|
602
|
|
603 static inline void write_back_intra_pred_mode(H264Context *h){
|
|
604 MpegEncContext * const s = &h->s;
|
|
605 const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
|
|
606
|
|
607 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
|
|
608 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
|
|
609 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
|
|
610 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
|
|
611 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
|
|
612 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
|
|
613 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
|
|
614 }
|
|
615
|
|
616 /**
|
|
617 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
|
|
618 */
|
|
619 static inline int check_intra4x4_pred_mode(H264Context *h){
|
|
620 MpegEncContext * const s = &h->s;
|
|
621 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
|
|
622 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
|
|
623 int i;
|
|
624
|
|
625 if(!(h->top_samples_available&0x8000)){
|
|
626 for(i=0; i<4; i++){
|
|
627 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
|
|
628 if(status<0){
|
|
629 fprintf(stderr, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
|
|
630 return -1;
|
|
631 } else if(status){
|
|
632 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
|
|
633 }
|
|
634 }
|
|
635 }
|
|
636
|
|
637 if(!(h->left_samples_available&0x8000)){
|
|
638 for(i=0; i<4; i++){
|
|
639 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
|
|
640 if(status<0){
|
|
641 fprintf(stderr, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
|
|
642 return -1;
|
|
643 } else if(status){
|
|
644 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
|
|
645 }
|
|
646 }
|
|
647 }
|
|
648
|
|
649 return 0;
|
|
650 } //FIXME cleanup like next
|
|
651
|
|
652 /**
|
|
653 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
|
|
654 */
|
|
655 static inline int check_intra_pred_mode(H264Context *h, int mode){
|
|
656 MpegEncContext * const s = &h->s;
|
|
657 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
|
|
658 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
|
|
659
|
|
660 if(!(h->top_samples_available&0x8000)){
|
|
661 mode= top[ mode ];
|
|
662 if(mode<0){
|
|
663 fprintf(stderr, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
|
|
664 return -1;
|
|
665 }
|
|
666 }
|
|
667
|
|
668 if(!(h->left_samples_available&0x8000)){
|
|
669 mode= left[ mode ];
|
|
670 if(mode<0){
|
|
671 fprintf(stderr, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
|
|
672 return -1;
|
|
673 }
|
|
674 }
|
|
675
|
|
676 return mode;
|
|
677 }
|
|
678
|
|
679 /**
|
|
680 * gets the predicted intra4x4 prediction mode.
|
|
681 */
|
|
682 static inline int pred_intra_mode(H264Context *h, int n){
|
|
683 const int index8= scan8[n];
|
|
684 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
|
|
685 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
|
|
686 const int min= FFMIN(left, top);
|
|
687
|
1170
|
688 tprintf("mode:%d %d min:%d\n", left ,top, min);
|
1168
|
689
|
|
690 if(min<0) return DC_PRED;
|
|
691 else return min;
|
|
692 }
|
|
693
|
|
694 static inline void write_back_non_zero_count(H264Context *h){
|
|
695 MpegEncContext * const s = &h->s;
|
|
696 const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
|
|
697
|
|
698 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
|
|
699 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
|
|
700 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
|
|
701 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
|
|
702 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
|
|
703 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
|
|
704 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
|
|
705
|
|
706 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
|
|
707 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
|
|
708 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
|
|
709
|
|
710 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
|
|
711 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
|
|
712 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
|
|
713 }
|
|
714
|
|
715 /**
|
|
716 * gets the predicted number of non zero coefficients.
|
|
717 * @param n block index
|
|
718 */
|
|
719 static inline int pred_non_zero_count(H264Context *h, int n){
|
|
720 const int index8= scan8[n];
|
|
721 const int left= h->non_zero_count_cache[index8 - 1];
|
|
722 const int top = h->non_zero_count_cache[index8 - 8];
|
|
723 int i= left + top;
|
|
724
|
|
725 if(i<64) i= (i+1)>>1;
|
|
726
|
1170
|
727 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
|
1168
|
728
|
|
729 return i&31;
|
|
730 }
|
|
731
|
1169
|
732 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
|
|
733 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
|
|
734
|
|
735 if(topright_ref != PART_NOT_AVAILABLE){
|
|
736 *C= h->mv_cache[list][ i - 8 + part_width ];
|
|
737 return topright_ref;
|
|
738 }else{
|
1170
|
739 tprintf("topright MV not available\n");
|
|
740
|
1169
|
741 *C= h->mv_cache[list][ i - 8 - 1 ];
|
|
742 return h->ref_cache[list][ i - 8 - 1 ];
|
|
743 }
|
|
744 }
|
|
745
|
1168
|
746 /**
|
|
747 * gets the predicted MV.
|
|
748 * @param n the block index
|
|
749 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
|
|
750 * @param mx the x component of the predicted motion vector
|
|
751 * @param my the y component of the predicted motion vector
|
|
752 */
|
|
753 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
|
|
754 MpegEncContext * const s = &h->s;
|
|
755 const int index8= scan8[n];
|
|
756 const int top_ref= h->ref_cache[list][ index8 - 8 ];
|
|
757 const int left_ref= h->ref_cache[list][ index8 - 1 ];
|
|
758 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
|
|
759 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
|
1169
|
760 const int16_t * C;
|
|
761 int diagonal_ref, match_count;
|
|
762
|
1168
|
763 assert(part_width==1 || part_width==2 || part_width==4);
|
1169
|
764
|
1168
|
765 /* mv_cache
|
|
766 B . . A T T T T
|
|
767 U . . L . . , .
|
|
768 U . . L . . . .
|
|
769 U . . L . . , .
|
|
770 . . . L . . . .
|
|
771 */
|
1169
|
772
|
|
773 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
|
|
774 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
|
|
775
|
|
776 if(match_count > 1){ //most common
|
|
777 *mx= mid_pred(A[0], B[0], C[0]);
|
|
778 *my= mid_pred(A[1], B[1], C[1]);
|
|
779 }else if(match_count==1){
|
|
780 if(left_ref==ref){
|
|
781 *mx= A[0];
|
|
782 *my= A[1];
|
|
783 }else if(top_ref==ref){
|
|
784 *mx= B[0];
|
|
785 *my= B[1];
|
|
786 }else{
|
|
787 *mx= C[0];
|
|
788 *my= C[1];
|
|
789 }
|
|
790 }else{
|
|
791 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
|
|
792 *mx= A[0];
|
|
793 *my= A[1];
|
1168
|
794 }else{
|
|
795 *mx= mid_pred(A[0], B[0], C[0]);
|
|
796 *my= mid_pred(A[1], B[1], C[1]);
|
|
797 }
|
1169
|
798 }
|
1168
|
799
|
1170
|
800 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, s->mb_x, s->mb_y, n, list);
|
1168
|
801 }
|
|
802
|
|
803 /**
|
|
804 * gets the directionally predicted 16x8 MV.
|
|
805 * @param n the block index
|
|
806 * @param mx the x component of the predicted motion vector
|
|
807 * @param my the y component of the predicted motion vector
|
|
808 */
|
|
809 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
|
|
810 MpegEncContext * const s = &h->s;
|
|
811 if(n==0){
|
|
812 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
|
|
813 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
|
|
814
|
1170
|
815 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", top_ref, B[0], B[1], s->mb_x, s->mb_y, n, list);
|
1168
|
816
|
|
817 if(top_ref == ref){
|
|
818 *mx= B[0];
|
|
819 *my= B[1];
|
|
820 return;
|
|
821 }
|
|
822 }else{
|
|
823 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
|
|
824 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
|
|
825
|
1170
|
826 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], s->mb_x, s->mb_y, n, list);
|
1168
|
827
|
|
828 if(left_ref == ref){
|
|
829 *mx= A[0];
|
|
830 *my= A[1];
|
|
831 return;
|
|
832 }
|
|
833 }
|
|
834
|
|
835 //RARE
|
|
836 pred_motion(h, n, 4, list, ref, mx, my);
|
|
837 }
|
|
838
|
|
839 /**
|
|
840 * gets the directionally predicted 8x16 MV.
|
|
841 * @param n the block index
|
|
842 * @param mx the x component of the predicted motion vector
|
|
843 * @param my the y component of the predicted motion vector
|
|
844 */
|
|
845 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
|
|
846 MpegEncContext * const s = &h->s;
|
|
847 if(n==0){
|
|
848 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
|
|
849 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
|
|
850
|
1170
|
851 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], s->mb_x, s->mb_y, n, list);
|
1168
|
852
|
|
853 if(left_ref == ref){
|
|
854 *mx= A[0];
|
|
855 *my= A[1];
|
|
856 return;
|
|
857 }
|
|
858 }else{
|
1169
|
859 const int16_t * C;
|
|
860 int diagonal_ref;
|
|
861
|
|
862 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
|
1168
|
863
|
1170
|
864 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", diagonal_ref, C[0], C[1], s->mb_x, s->mb_y, n, list);
|
1168
|
865
|
1169
|
866 if(diagonal_ref == ref){
|
1168
|
867 *mx= C[0];
|
|
868 *my= C[1];
|
|
869 return;
|
|
870 }
|
|
871 }
|
|
872
|
|
873 //RARE
|
|
874 pred_motion(h, n, 2, list, ref, mx, my);
|
|
875 }
|
|
876
|
|
877 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
|
|
878 MpegEncContext * const s = &h->s;
|
|
879 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
|
|
880 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
|
|
881
|
1170
|
882 tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, s->mb_x, s->mb_y);
|
1168
|
883
|
|
884 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
|
|
885 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
|
|
886 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
|
|
887
|
|
888 *mx = *my = 0;
|
|
889 return;
|
|
890 }
|
|
891
|
|
892 pred_motion(h, 0, 4, 0, 0, mx, my);
|
|
893
|
|
894 return;
|
|
895 }
|
|
896
|
|
897 static inline void write_back_motion(H264Context *h, int mb_type){
|
|
898 MpegEncContext * const s = &h->s;
|
|
899 const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
|
|
900 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
|
|
901 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
|
|
902 int list;
|
|
903
|
|
904 for(list=0; list<2; list++){
|
|
905 int y;
|
|
906 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
|
|
907 if(1){ //FIXME skip or never read if mb_type doesnt use it
|
|
908 for(y=0; y<4; y++){
|
|
909 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
|
|
910 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
|
|
911 }
|
|
912 for(y=0; y<2; y++){
|
|
913 *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
|
|
914 }
|
|
915 }
|
|
916 continue; //FIXME direct mode ...
|
|
917 }
|
|
918
|
|
919 for(y=0; y<4; y++){
|
|
920 *(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];
|
|
921 *(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];
|
|
922 }
|
|
923 for(y=0; y<2; y++){
|
|
924 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
|
|
925 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
|
|
926 }
|
|
927 }
|
|
928 }
|
|
929
|
|
930 /**
|
|
931 * Decodes a network abstraction layer unit.
|
|
932 * @param consumed is the number of bytes used as input
|
|
933 * @param length is the length of the array
|
|
934 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
|
|
935 * @returns decoded bytes, might be src+1 if no escapes
|
|
936 */
|
|
937 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
|
|
938 int i, si, di;
|
|
939 uint8_t *dst;
|
|
940
|
|
941 // src[0]&0x80; //forbidden bit
|
|
942 h->nal_ref_idc= src[0]>>5;
|
|
943 h->nal_unit_type= src[0]&0x1F;
|
|
944
|
|
945 src++; length--;
|
|
946 #if 0
|
|
947 for(i=0; i<length; i++)
|
|
948 printf("%2X ", src[i]);
|
|
949 #endif
|
|
950 for(i=0; i+1<length; i+=2){
|
|
951 if(src[i]) continue;
|
|
952 if(i>0 && src[i-1]==0) i--;
|
|
953 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
|
|
954 if(src[i+2]!=3){
|
|
955 /* startcode, so we must be past the end */
|
|
956 length=i;
|
|
957 }
|
|
958 break;
|
|
959 }
|
|
960 }
|
|
961
|
|
962 if(i>=length-1){ //no escaped 0
|
|
963 *dst_length= length;
|
|
964 *consumed= length+1; //+1 for the header
|
|
965 return src;
|
|
966 }
|
|
967
|
|
968 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
|
|
969 dst= h->rbsp_buffer;
|
|
970
|
|
971 //printf("deoding esc\n");
|
|
972 si=di=0;
|
|
973 while(si<length){
|
|
974 //remove escapes (very rare 1:2^22)
|
|
975 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
|
|
976 if(src[si+2]==3){ //escape
|
|
977 dst[di++]= 0;
|
|
978 dst[di++]= 0;
|
|
979 si+=3;
|
|
980 }else //next start code
|
|
981 break;
|
|
982 }
|
|
983
|
|
984 dst[di++]= src[si++];
|
|
985 }
|
|
986
|
|
987 *dst_length= di;
|
|
988 *consumed= si + 1;//+1 for the header
|
|
989 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
|
|
990 return dst;
|
|
991 }
|
|
992
|
|
993 /**
|
|
994 * @param src the data which should be escaped
|
|
995 * @param dst the target buffer, dst+1 == src is allowed as a special case
|
|
996 * @param length the length of the src data
|
|
997 * @param dst_length the length of the dst array
|
|
998 * @returns length of escaped data in bytes or -1 if an error occured
|
|
999 */
|
|
1000 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
|
|
1001 int i, escape_count, si, di;
|
|
1002 uint8_t *temp;
|
|
1003
|
|
1004 assert(length>=0);
|
|
1005 assert(dst_length>0);
|
|
1006
|
|
1007 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
|
|
1008
|
|
1009 if(length==0) return 1;
|
|
1010
|
|
1011 escape_count= 0;
|
|
1012 for(i=0; i<length; i+=2){
|
|
1013 if(src[i]) continue;
|
|
1014 if(i>0 && src[i-1]==0)
|
|
1015 i--;
|
|
1016 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
|
|
1017 escape_count++;
|
|
1018 i+=2;
|
|
1019 }
|
|
1020 }
|
|
1021
|
|
1022 if(escape_count==0){
|
|
1023 if(dst+1 != src)
|
|
1024 memcpy(dst+1, src, length);
|
|
1025 return length + 1;
|
|
1026 }
|
|
1027
|
|
1028 if(length + escape_count + 1> dst_length)
|
|
1029 return -1;
|
|
1030
|
|
1031 //this should be damn rare (hopefully)
|
|
1032
|
|
1033 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
|
|
1034 temp= h->rbsp_buffer;
|
|
1035 //printf("encoding esc\n");
|
|
1036
|
|
1037 si= 0;
|
|
1038 di= 0;
|
|
1039 while(si < length){
|
|
1040 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
|
|
1041 temp[di++]= 0; si++;
|
|
1042 temp[di++]= 0; si++;
|
|
1043 temp[di++]= 3;
|
|
1044 temp[di++]= src[si++];
|
|
1045 }
|
|
1046 else
|
|
1047 temp[di++]= src[si++];
|
|
1048 }
|
|
1049 memcpy(dst+1, temp, length+escape_count);
|
|
1050
|
|
1051 assert(di == length+escape_count);
|
|
1052
|
|
1053 return di + 1;
|
|
1054 }
|
|
1055
|
|
1056 /**
|
|
1057 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
|
|
1058 */
|
|
1059 static void encode_rbsp_trailing(PutBitContext *pb){
|
|
1060 int length;
|
|
1061 put_bits(pb, 1, 1);
|
|
1062 length= (-get_bit_count(pb))&7;
|
|
1063 if(length) put_bits(pb, length, 0);
|
|
1064 }
|
|
1065
|
|
1066 /**
|
|
1067 * identifies the exact end of the bitstream
|
|
1068 * @return the length of the trailing, or 0 if damaged
|
|
1069 */
|
|
1070 static int decode_rbsp_trailing(uint8_t *src){
|
|
1071 int v= *src;
|
|
1072 int r;
|
|
1073
|
1170
|
1074 tprintf("rbsp trailing %X\n", v);
|
1168
|
1075
|
|
1076 for(r=1; r<9; r++){
|
|
1077 if(v&1) return r;
|
|
1078 v>>=1;
|
|
1079 }
|
|
1080 return 0;
|
|
1081 }
|
|
1082
|
|
1083 /**
|
|
1084 * idct tranforms the 16 dc values and dequantize them.
|
|
1085 * @param qp quantization parameter
|
|
1086 */
|
|
1087 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
|
|
1088 const int qmul= dequant_coeff[qp][0];
|
|
1089 #define stride 16
|
|
1090 int i;
|
|
1091 int temp[16]; //FIXME check if this is a good idea
|
|
1092 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
|
|
1093 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
|
|
1094
|
|
1095 //memset(block, 64, 2*256);
|
|
1096 //return;
|
|
1097 for(i=0; i<4; i++){
|
|
1098 const int offset= y_offset[i];
|
|
1099 const int z0= block[offset+stride*0] + block[offset+stride*4];
|
|
1100 const int z1= block[offset+stride*0] - block[offset+stride*4];
|
|
1101 const int z2= block[offset+stride*1] - block[offset+stride*5];
|
|
1102 const int z3= block[offset+stride*1] + block[offset+stride*5];
|
|
1103
|
|
1104 temp[4*i+0]= z0+z3;
|
|
1105 temp[4*i+1]= z1+z2;
|
|
1106 temp[4*i+2]= z1-z2;
|
|
1107 temp[4*i+3]= z0-z3;
|
|
1108 }
|
|
1109
|
|
1110 for(i=0; i<4; i++){
|
|
1111 const int offset= x_offset[i];
|
|
1112 const int z0= temp[4*0+i] + temp[4*2+i];
|
|
1113 const int z1= temp[4*0+i] - temp[4*2+i];
|
|
1114 const int z2= temp[4*1+i] - temp[4*3+i];
|
|
1115 const int z3= temp[4*1+i] + temp[4*3+i];
|
|
1116
|
|
1117 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
|
|
1118 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
|
|
1119 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
|
|
1120 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
|
|
1121 }
|
|
1122 }
|
|
1123
|
|
1124 /**
|
|
1125 * dct tranforms the 16 dc values.
|
|
1126 * @param qp quantization parameter ??? FIXME
|
|
1127 */
|
|
1128 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
|
|
1129 // const int qmul= dequant_coeff[qp][0];
|
|
1130 int i;
|
|
1131 int temp[16]; //FIXME check if this is a good idea
|
|
1132 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
|
|
1133 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
|
|
1134
|
|
1135 for(i=0; i<4; i++){
|
|
1136 const int offset= y_offset[i];
|
|
1137 const int z0= block[offset+stride*0] + block[offset+stride*4];
|
|
1138 const int z1= block[offset+stride*0] - block[offset+stride*4];
|
|
1139 const int z2= block[offset+stride*1] - block[offset+stride*5];
|
|
1140 const int z3= block[offset+stride*1] + block[offset+stride*5];
|
|
1141
|
|
1142 temp[4*i+0]= z0+z3;
|
|
1143 temp[4*i+1]= z1+z2;
|
|
1144 temp[4*i+2]= z1-z2;
|
|
1145 temp[4*i+3]= z0-z3;
|
|
1146 }
|
|
1147
|
|
1148 for(i=0; i<4; i++){
|
|
1149 const int offset= x_offset[i];
|
|
1150 const int z0= temp[4*0+i] + temp[4*2+i];
|
|
1151 const int z1= temp[4*0+i] - temp[4*2+i];
|
|
1152 const int z2= temp[4*1+i] - temp[4*3+i];
|
|
1153 const int z3= temp[4*1+i] + temp[4*3+i];
|
|
1154
|
|
1155 block[stride*0 +offset]= (z0 + z3)>>1;
|
|
1156 block[stride*2 +offset]= (z1 + z2)>>1;
|
|
1157 block[stride*8 +offset]= (z1 - z2)>>1;
|
|
1158 block[stride*10+offset]= (z0 - z3)>>1;
|
|
1159 }
|
|
1160 }
|
|
1161 #undef xStride
|
|
1162 #undef stride
|
|
1163
|
|
1164 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
|
|
1165 const int qmul= dequant_coeff[qp][0];
|
|
1166 const int stride= 16*2;
|
|
1167 const int xStride= 16;
|
|
1168 int a,b,c,d,e;
|
|
1169
|
|
1170 a= block[stride*0 + xStride*0];
|
|
1171 b= block[stride*0 + xStride*1];
|
|
1172 c= block[stride*1 + xStride*0];
|
|
1173 d= block[stride*1 + xStride*1];
|
|
1174
|
|
1175 e= a-b;
|
|
1176 a= a+b;
|
|
1177 b= c-d;
|
|
1178 c= c+d;
|
|
1179
|
|
1180 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
|
|
1181 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
|
|
1182 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
|
|
1183 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
|
|
1184 }
|
|
1185
|
|
1186 static void chroma_dc_dct_c(DCTELEM *block){
|
|
1187 const int stride= 16*2;
|
|
1188 const int xStride= 16;
|
|
1189 int a,b,c,d,e;
|
|
1190
|
|
1191 a= block[stride*0 + xStride*0];
|
|
1192 b= block[stride*0 + xStride*1];
|
|
1193 c= block[stride*1 + xStride*0];
|
|
1194 d= block[stride*1 + xStride*1];
|
|
1195
|
|
1196 e= a-b;
|
|
1197 a= a+b;
|
|
1198 b= c-d;
|
|
1199 c= c+d;
|
|
1200
|
|
1201 block[stride*0 + xStride*0]= (a+c);
|
|
1202 block[stride*0 + xStride*1]= (e+b);
|
|
1203 block[stride*1 + xStride*0]= (a-c);
|
|
1204 block[stride*1 + xStride*1]= (e-b);
|
|
1205 }
|
|
1206
|
|
1207 /**
|
|
1208 * gets the chroma qp.
|
|
1209 */
|
|
1210 static inline int get_chroma_qp(H264Context *h, int qscale){
|
|
1211
|
|
1212 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
|
|
1213 }
|
|
1214
|
|
1215
|
|
1216 /**
|
|
1217 *
|
|
1218 */
|
|
1219 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
|
|
1220 int i;
|
|
1221 uint8_t *cm = cropTbl + MAX_NEG_CROP;
|
|
1222
|
|
1223 block[0] += 32;
|
|
1224 #if 1
|
|
1225 for(i=0; i<4; i++){
|
|
1226 const int z0= block[i + 4*0] + block[i + 4*2];
|
|
1227 const int z1= block[i + 4*0] - block[i + 4*2];
|
|
1228 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
|
|
1229 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
|
|
1230
|
|
1231 block[i + 4*0]= z0 + z3;
|
|
1232 block[i + 4*1]= z1 + z2;
|
|
1233 block[i + 4*2]= z1 - z2;
|
|
1234 block[i + 4*3]= z0 - z3;
|
|
1235 }
|
|
1236
|
|
1237 for(i=0; i<4; i++){
|
|
1238 const int z0= block[0 + 4*i] + block[2 + 4*i];
|
|
1239 const int z1= block[0 + 4*i] - block[2 + 4*i];
|
|
1240 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i];
|
|
1241 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1);
|
|
1242
|
|
1243 dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
|
|
1244 dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
|
|
1245 dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
|
|
1246 dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
|
|
1247 }
|
|
1248 #else
|
|
1249 for(i=0; i<4; i++){
|
|
1250 const int z0= block[0 + 4*i] + block[2 + 4*i];
|
|
1251 const int z1= block[0 + 4*i] - block[2 + 4*i];
|
|
1252 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i];
|
|
1253 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1);
|
|
1254
|
|
1255 block[0 + 4*i]= z0 + z3;
|
|
1256 block[1 + 4*i]= z1 + z2;
|
|
1257 block[2 + 4*i]= z1 - z2;
|
|
1258 block[3 + 4*i]= z0 - z3;
|
|
1259 }
|
|
1260
|
|
1261 for(i=0; i<4; i++){
|
|
1262 const int z0= block[i + 4*0] + block[i + 4*2];
|
|
1263 const int z1= block[i + 4*0] - block[i + 4*2];
|
|
1264 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
|
|
1265 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
|
|
1266
|
|
1267 dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
|
|
1268 dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
|
|
1269 dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
|
|
1270 dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
|
|
1271 }
|
|
1272 #endif
|
|
1273 }
|
|
1274
|
|
1275 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
|
|
1276 int i;
|
|
1277 //FIXME try int temp instead of block
|
|
1278
|
|
1279 for(i=0; i<4; i++){
|
|
1280 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
|
|
1281 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
|
|
1282 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
|
|
1283 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
|
|
1284 const int z0= d0 + d3;
|
|
1285 const int z3= d0 - d3;
|
|
1286 const int z1= d1 + d2;
|
|
1287 const int z2= d1 - d2;
|
|
1288
|
|
1289 block[0 + 4*i]= z0 + z1;
|
|
1290 block[1 + 4*i]= 2*z3 + z2;
|
|
1291 block[2 + 4*i]= z0 - z1;
|
|
1292 block[3 + 4*i]= z3 - 2*z2;
|
|
1293 }
|
|
1294
|
|
1295 for(i=0; i<4; i++){
|
|
1296 const int z0= block[0*4 + i] + block[3*4 + i];
|
|
1297 const int z3= block[0*4 + i] - block[3*4 + i];
|
|
1298 const int z1= block[1*4 + i] + block[2*4 + i];
|
|
1299 const int z2= block[1*4 + i] - block[2*4 + i];
|
|
1300
|
|
1301 block[0*4 + i]= z0 + z1;
|
|
1302 block[1*4 + i]= 2*z3 + z2;
|
|
1303 block[2*4 + i]= z0 - z1;
|
|
1304 block[3*4 + i]= z3 - 2*z2;
|
|
1305 }
|
|
1306 }
|
|
1307
|
|
1308 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
|
|
1309 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
|
|
1310 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
|
|
1311 int i;
|
|
1312 const int * const quant_table= quant_coeff[qscale];
|
|
1313 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
|
|
1314 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
|
|
1315 const unsigned int threshold2= (threshold1<<1);
|
|
1316 int last_non_zero;
|
|
1317
|
|
1318 if(seperate_dc){
|
|
1319 if(qscale<=18){
|
|
1320 //avoid overflows
|
|
1321 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
|
|
1322 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
|
|
1323 const unsigned int dc_threshold2= (dc_threshold1<<1);
|
|
1324
|
|
1325 int level= block[0]*quant_coeff[qscale+18][0];
|
|
1326 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
|
|
1327 if(level>0){
|
|
1328 level= (dc_bias + level)>>(QUANT_SHIFT-2);
|
|
1329 block[0]= level;
|
|
1330 }else{
|
|
1331 level= (dc_bias - level)>>(QUANT_SHIFT-2);
|
|
1332 block[0]= -level;
|
|
1333 }
|
|
1334 // last_non_zero = i;
|
|
1335 }else{
|
|
1336 block[0]=0;
|
|
1337 }
|
|
1338 }else{
|
|
1339 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
|
|
1340 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
|
|
1341 const unsigned int dc_threshold2= (dc_threshold1<<1);
|
|
1342
|
|
1343 int level= block[0]*quant_table[0];
|
|
1344 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
|
|
1345 if(level>0){
|
|
1346 level= (dc_bias + level)>>(QUANT_SHIFT+1);
|
|
1347 block[0]= level;
|
|
1348 }else{
|
|
1349 level= (dc_bias - level)>>(QUANT_SHIFT+1);
|
|
1350 block[0]= -level;
|
|
1351 }
|
|
1352 // last_non_zero = i;
|
|
1353 }else{
|
|
1354 block[0]=0;
|
|
1355 }
|
|
1356 }
|
|
1357 last_non_zero= 0;
|
|
1358 i=1;
|
|
1359 }else{
|
|
1360 last_non_zero= -1;
|
|
1361 i=0;
|
|
1362 }
|
|
1363
|
|
1364 for(; i<16; i++){
|
|
1365 const int j= scantable[i];
|
|
1366 int level= block[j]*quant_table[j];
|
|
1367
|
|
1368 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
|
|
1369 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
|
|
1370 if(((unsigned)(level+threshold1))>threshold2){
|
|
1371 if(level>0){
|
|
1372 level= (bias + level)>>QUANT_SHIFT;
|
|
1373 block[j]= level;
|
|
1374 }else{
|
|
1375 level= (bias - level)>>QUANT_SHIFT;
|
|
1376 block[j]= -level;
|
|
1377 }
|
|
1378 last_non_zero = i;
|
|
1379 }else{
|
|
1380 block[j]=0;
|
|
1381 }
|
|
1382 }
|
|
1383
|
|
1384 return last_non_zero;
|
|
1385 }
|
|
1386
|
|
1387 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
|
|
1388 const uint32_t a= ((uint32_t*)(src-stride))[0];
|
|
1389 ((uint32_t*)(src+0*stride))[0]= a;
|
|
1390 ((uint32_t*)(src+1*stride))[0]= a;
|
|
1391 ((uint32_t*)(src+2*stride))[0]= a;
|
|
1392 ((uint32_t*)(src+3*stride))[0]= a;
|
|
1393 }
|
|
1394
|
|
1395 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
|
|
1396 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
|
|
1397 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
|
|
1398 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
|
|
1399 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
|
|
1400 }
|
|
1401
|
|
1402 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
|
|
1403 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
|
|
1404 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
|
|
1405
|
|
1406 ((uint32_t*)(src+0*stride))[0]=
|
|
1407 ((uint32_t*)(src+1*stride))[0]=
|
|
1408 ((uint32_t*)(src+2*stride))[0]=
|
|
1409 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
|
|
1410 }
|
|
1411
|
|
1412 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
|
|
1413 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
|
|
1414
|
|
1415 ((uint32_t*)(src+0*stride))[0]=
|
|
1416 ((uint32_t*)(src+1*stride))[0]=
|
|
1417 ((uint32_t*)(src+2*stride))[0]=
|
|
1418 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
|
|
1419 }
|
|
1420
|
|
1421 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
|
|
1422 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
|
|
1423
|
|
1424 ((uint32_t*)(src+0*stride))[0]=
|
|
1425 ((uint32_t*)(src+1*stride))[0]=
|
|
1426 ((uint32_t*)(src+2*stride))[0]=
|
|
1427 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
|
|
1428 }
|
|
1429
|
|
1430 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
|
|
1431 ((uint32_t*)(src+0*stride))[0]=
|
|
1432 ((uint32_t*)(src+1*stride))[0]=
|
|
1433 ((uint32_t*)(src+2*stride))[0]=
|
|
1434 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
|
|
1435 }
|
|
1436
|
|
1437
|
|
1438 #define LOAD_TOP_RIGHT_EDGE\
|
|
1439 const int t4= topright[0];\
|
|
1440 const int t5= topright[1];\
|
|
1441 const int t6= topright[2];\
|
|
1442 const int t7= topright[3];\
|
|
1443
|
|
1444 #define LOAD_LEFT_EDGE\
|
|
1445 const int l0= src[-1+0*stride];\
|
|
1446 const int l1= src[-1+1*stride];\
|
|
1447 const int l2= src[-1+2*stride];\
|
|
1448 const int l3= src[-1+3*stride];\
|
|
1449
|
|
1450 #define LOAD_TOP_EDGE\
|
|
1451 const int t0= src[ 0-1*stride];\
|
|
1452 const int t1= src[ 1-1*stride];\
|
|
1453 const int t2= src[ 2-1*stride];\
|
|
1454 const int t3= src[ 3-1*stride];\
|
|
1455
|
|
1456 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
|
|
1457 const int lt= src[-1-1*stride];
|
|
1458 LOAD_TOP_EDGE
|
|
1459 LOAD_LEFT_EDGE
|
|
1460
|
|
1461 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
|
|
1462 src[0+2*stride]=
|
|
1463 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
|
|
1464 src[0+1*stride]=
|
|
1465 src[1+2*stride]=
|
|
1466 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
|
|
1467 src[0+0*stride]=
|
|
1468 src[1+1*stride]=
|
|
1469 src[2+2*stride]=
|
|
1470 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
|
|
1471 src[1+0*stride]=
|
|
1472 src[2+1*stride]=
|
|
1473 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
|
|
1474 src[2+0*stride]=
|
|
1475 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
|
|
1476 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
|
|
1477 };
|
|
1478
|
|
1479 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
|
|
1480 LOAD_TOP_EDGE
|
|
1481 LOAD_TOP_RIGHT_EDGE
|
|
1482 // LOAD_LEFT_EDGE
|
|
1483
|
|
1484 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
|
|
1485 src[1+0*stride]=
|
|
1486 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
|
|
1487 src[2+0*stride]=
|
|
1488 src[1+1*stride]=
|
|
1489 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
|
|
1490 src[3+0*stride]=
|
|
1491 src[2+1*stride]=
|
|
1492 src[1+2*stride]=
|
|
1493 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
|
|
1494 src[3+1*stride]=
|
|
1495 src[2+2*stride]=
|
|
1496 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
|
|
1497 src[3+2*stride]=
|
|
1498 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
|
|
1499 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
|
|
1500 };
|
|
1501
|
|
1502 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
|
|
1503 const int lt= src[-1-1*stride];
|
|
1504 LOAD_TOP_EDGE
|
|
1505 LOAD_LEFT_EDGE
|
|
1506 const __attribute__((unused)) int unu= l3;
|
|
1507
|
|
1508 src[0+0*stride]=
|
|
1509 src[1+2*stride]=(lt + t0 + 1)>>1;
|
|
1510 src[1+0*stride]=
|
|
1511 src[2+2*stride]=(t0 + t1 + 1)>>1;
|
|
1512 src[2+0*stride]=
|
|
1513 src[3+2*stride]=(t1 + t2 + 1)>>1;
|
|
1514 src[3+0*stride]=(t2 + t3 + 1)>>1;
|
|
1515 src[0+1*stride]=
|
|
1516 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
|
|
1517 src[1+1*stride]=
|
|
1518 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
|
|
1519 src[2+1*stride]=
|
|
1520 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
|
|
1521 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
|
|
1522 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
|
|
1523 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
|
|
1524 };
|
|
1525
|
|
1526 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
|
|
1527 LOAD_TOP_EDGE
|
|
1528 LOAD_TOP_RIGHT_EDGE
|
|
1529 const __attribute__((unused)) int unu= t7;
|
|
1530
|
|
1531 src[0+0*stride]=(t0 + t1 + 1)>>1;
|
|
1532 src[1+0*stride]=
|
|
1533 src[0+2*stride]=(t1 + t2 + 1)>>1;
|
|
1534 src[2+0*stride]=
|
|
1535 src[1+2*stride]=(t2 + t3 + 1)>>1;
|
|
1536 src[3+0*stride]=
|
|
1537 src[2+2*stride]=(t3 + t4+ 1)>>1;
|
|
1538 src[3+2*stride]=(t4 + t5+ 1)>>1;
|
|
1539 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
|
|
1540 src[1+1*stride]=
|
|
1541 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
|
|
1542 src[2+1*stride]=
|
|
1543 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
|
|
1544 src[3+1*stride]=
|
|
1545 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
|
|
1546 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
|
|
1547 };
|
|
1548
|
|
1549 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
|
|
1550 LOAD_LEFT_EDGE
|
|
1551
|
|
1552 src[0+0*stride]=(l0 + l1 + 1)>>1;
|
|
1553 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
|
|
1554 src[2+0*stride]=
|
|
1555 src[0+1*stride]=(l1 + l2 + 1)>>1;
|
|
1556 src[3+0*stride]=
|
|
1557 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
|
|
1558 src[2+1*stride]=
|
|
1559 src[0+2*stride]=(l2 + l3 + 1)>>1;
|
|
1560 src[3+1*stride]=
|
|
1561 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
|
|
1562 src[3+2*stride]=
|
|
1563 src[1+3*stride]=
|
|
1564 src[0+3*stride]=
|
|
1565 src[2+2*stride]=
|
|
1566 src[2+3*stride]=
|
|
1567 src[3+3*stride]=l3;
|
|
1568 };
|
|
1569
|
|
1570 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
|
|
1571 const int lt= src[-1-1*stride];
|
|
1572 LOAD_TOP_EDGE
|
|
1573 LOAD_LEFT_EDGE
|
|
1574 const __attribute__((unused)) int unu= t3;
|
|
1575
|
|
1576 src[0+0*stride]=
|
|
1577 src[2+1*stride]=(lt + l0 + 1)>>1;
|
|
1578 src[1+0*stride]=
|
|
1579 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
|
|
1580 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
|
|
1581 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
|
|
1582 src[0+1*stride]=
|
|
1583 src[2+2*stride]=(l0 + l1 + 1)>>1;
|
|
1584 src[1+1*stride]=
|
|
1585 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
|
|
1586 src[0+2*stride]=
|
|
1587 src[2+3*stride]=(l1 + l2+ 1)>>1;
|
|
1588 src[1+2*stride]=
|
|
1589 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
|
|
1590 src[0+3*stride]=(l2 + l3 + 1)>>1;
|
|
1591 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
|
|
1592 };
|
|
1593
|
|
1594 static void pred16x16_vertical_c(uint8_t *src, int stride){
|
|
1595 int i;
|
|
1596 const uint32_t a= ((uint32_t*)(src-stride))[0];
|
|
1597 const uint32_t b= ((uint32_t*)(src-stride))[1];
|
|
1598 const uint32_t c= ((uint32_t*)(src-stride))[2];
|
|
1599 const uint32_t d= ((uint32_t*)(src-stride))[3];
|
|
1600
|
|
1601 for(i=0; i<16; i++){
|
|
1602 ((uint32_t*)(src+i*stride))[0]= a;
|
|
1603 ((uint32_t*)(src+i*stride))[1]= b;
|
|
1604 ((uint32_t*)(src+i*stride))[2]= c;
|
|
1605 ((uint32_t*)(src+i*stride))[3]= d;
|
|
1606 }
|
|
1607 }
|
|
1608
|
|
1609 static void pred16x16_horizontal_c(uint8_t *src, int stride){
|
|
1610 int i;
|
|
1611
|
|
1612 for(i=0; i<16; i++){
|
|
1613 ((uint32_t*)(src+i*stride))[0]=
|
|
1614 ((uint32_t*)(src+i*stride))[1]=
|
|
1615 ((uint32_t*)(src+i*stride))[2]=
|
|
1616 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
|
|
1617 }
|
|
1618 }
|
|
1619
|
|
1620 static void pred16x16_dc_c(uint8_t *src, int stride){
|
|
1621 int i, dc=0;
|
|
1622
|
|
1623 for(i=0;i<16; i++){
|
|
1624 dc+= src[-1+i*stride];
|
|
1625 }
|
|
1626
|
|
1627 for(i=0;i<16; i++){
|
|
1628 dc+= src[i-stride];
|
|
1629 }
|
|
1630
|
|
1631 dc= 0x01010101*((dc + 16)>>5);
|
|
1632
|
|
1633 for(i=0; i<16; i++){
|
|
1634 ((uint32_t*)(src+i*stride))[0]=
|
|
1635 ((uint32_t*)(src+i*stride))[1]=
|
|
1636 ((uint32_t*)(src+i*stride))[2]=
|
|
1637 ((uint32_t*)(src+i*stride))[3]= dc;
|
|
1638 }
|
|
1639 }
|
|
1640
|
|
1641 static void pred16x16_left_dc_c(uint8_t *src, int stride){
|
|
1642 int i, dc=0;
|
|
1643
|
|
1644 for(i=0;i<16; i++){
|
|
1645 dc+= src[-1+i*stride];
|
|
1646 }
|
|
1647
|
|
1648 dc= 0x01010101*((dc + 8)>>4);
|
|
1649
|
|
1650 for(i=0; i<16; i++){
|
|
1651 ((uint32_t*)(src+i*stride))[0]=
|
|
1652 ((uint32_t*)(src+i*stride))[1]=
|
|
1653 ((uint32_t*)(src+i*stride))[2]=
|
|
1654 ((uint32_t*)(src+i*stride))[3]= dc;
|
|
1655 }
|
|
1656 }
|
|
1657
|
|
1658 static void pred16x16_top_dc_c(uint8_t *src, int stride){
|
|
1659 int i, dc=0;
|
|
1660
|
|
1661 for(i=0;i<16; i++){
|
|
1662 dc+= src[i-stride];
|
|
1663 }
|
|
1664 dc= 0x01010101*((dc + 8)>>4);
|
|
1665
|
|
1666 for(i=0; i<16; i++){
|
|
1667 ((uint32_t*)(src+i*stride))[0]=
|
|
1668 ((uint32_t*)(src+i*stride))[1]=
|
|
1669 ((uint32_t*)(src+i*stride))[2]=
|
|
1670 ((uint32_t*)(src+i*stride))[3]= dc;
|
|
1671 }
|
|
1672 }
|
|
1673
|
|
1674 static void pred16x16_128_dc_c(uint8_t *src, int stride){
|
|
1675 int i;
|
|
1676
|
|
1677 for(i=0; i<16; i++){
|
|
1678 ((uint32_t*)(src+i*stride))[0]=
|
|
1679 ((uint32_t*)(src+i*stride))[1]=
|
|
1680 ((uint32_t*)(src+i*stride))[2]=
|
|
1681 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
|
|
1682 }
|
|
1683 }
|
|
1684
|
|
1685 static void pred16x16_plane_c(uint8_t *src, int stride){
|
|
1686 uint8_t *cm = cropTbl + MAX_NEG_CROP;
|
|
1687 int i, dx, dy, dc;
|
|
1688 int temp[16];
|
|
1689
|
|
1690 dc= 16*(src[15-stride] + src[-1+15*stride]);
|
|
1691
|
|
1692 dx=dy=0;
|
|
1693 for(i=1; i<9; i++){
|
|
1694 dx += i*(src[7+i-stride] - src[7-i-stride]);
|
|
1695 dy += i*(src[-1+(7+i)*stride] - src[-1+(7-i)*stride]);
|
|
1696 }
|
|
1697 dx= (5*dx+32)>>6;
|
|
1698 dy= (5*dy+32)>>6;
|
|
1699
|
|
1700 dc += 16;
|
|
1701
|
|
1702 //FIXME modifiy dc,dx,dy to avoid -7
|
|
1703
|
|
1704 for(i=0; i<16; i++)
|
|
1705 temp[i]= dx*(i-7) + dc;
|
|
1706
|
|
1707 if( (dc - ABS(dx)*8 - ABS(dy)*8)>>5 < 0
|
|
1708 || (dc + ABS(dx)*8 + ABS(dy)*8)>>5 > 255){
|
|
1709
|
|
1710 for(i=0; i<16; i++){
|
|
1711 int j;
|
|
1712 for(j=0; j<16; j++)
|
|
1713 src[j + i*stride]= cm[ (temp[j] + dy*(i-7))>>5 ];
|
|
1714 }
|
|
1715 }else{
|
|
1716 for(i=0; i<16; i++){
|
|
1717 int j;
|
|
1718 for(j=0; j<16; j++)
|
|
1719 src[j + i*stride]= (temp[j] + dy*(i-7))>>5;
|
|
1720 }
|
|
1721 }
|
|
1722 }
|
|
1723
|
|
1724 static void pred8x8_vertical_c(uint8_t *src, int stride){
|
|
1725 int i;
|
|
1726 const uint32_t a= ((uint32_t*)(src-stride))[0];
|
|
1727 const uint32_t b= ((uint32_t*)(src-stride))[1];
|
|
1728
|
|
1729 for(i=0; i<8; i++){
|
|
1730 ((uint32_t*)(src+i*stride))[0]= a;
|
|
1731 ((uint32_t*)(src+i*stride))[1]= b;
|
|
1732 }
|
|
1733 }
|
|
1734
|
|
1735 static void pred8x8_horizontal_c(uint8_t *src, int stride){
|
|
1736 int i;
|
|
1737
|
|
1738 for(i=0; i<8; i++){
|
|
1739 ((uint32_t*)(src+i*stride))[0]=
|
|
1740 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
|
|
1741 }
|
|
1742 }
|
|
1743
|
|
1744 static void pred8x8_128_dc_c(uint8_t *src, int stride){
|
|
1745 int i;
|
|
1746
|
|
1747 for(i=0; i<4; i++){
|
|
1748 ((uint32_t*)(src+i*stride))[0]=
|
|
1749 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
|
|
1750 }
|
|
1751 for(i=4; i<8; i++){
|
|
1752 ((uint32_t*)(src+i*stride))[0]=
|
|
1753 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
|
|
1754 }
|
|
1755 }
|
|
1756
|
|
1757 static void pred8x8_left_dc_c(uint8_t *src, int stride){
|
|
1758 int i;
|
|
1759 int dc0, dc2;
|
|
1760
|
|
1761 dc0=dc2=0;
|
|
1762 for(i=0;i<4; i++){
|
|
1763 dc0+= src[-1+i*stride];
|
|
1764 dc2+= src[-1+(i+4)*stride];
|
|
1765 }
|
|
1766 dc0= 0x01010101*((dc0 + 2)>>2);
|
|
1767 dc2= 0x01010101*((dc2 + 2)>>2);
|
|
1768
|
|
1769 for(i=0; i<4; i++){
|
|
1770 ((uint32_t*)(src+i*stride))[0]=
|
|
1771 ((uint32_t*)(src+i*stride))[1]= dc0;
|
|
1772 }
|
|
1773 for(i=4; i<8; i++){
|
|
1774 ((uint32_t*)(src+i*stride))[0]=
|
|
1775 ((uint32_t*)(src+i*stride))[1]= dc2;
|
|
1776 }
|
|
1777 }
|
|
1778
|
|
1779 static void pred8x8_top_dc_c(uint8_t *src, int stride){
|
|
1780 int i;
|
|
1781 int dc0, dc1;
|
|
1782
|
|
1783 dc0=dc1=0;
|
|
1784 for(i=0;i<4; i++){
|
|
1785 dc0+= src[i-stride];
|
|
1786 dc1+= src[4+i-stride];
|
|
1787 }
|
|
1788 dc0= 0x01010101*((dc0 + 2)>>2);
|
|
1789 dc1= 0x01010101*((dc1 + 2)>>2);
|
|
1790
|
|
1791 for(i=0; i<4; i++){
|
|
1792 ((uint32_t*)(src+i*stride))[0]= dc0;
|
|
1793 ((uint32_t*)(src+i*stride))[1]= dc1;
|
|
1794 }
|
|
1795 for(i=4; i<8; i++){
|
|
1796 ((uint32_t*)(src+i*stride))[0]= dc0;
|
|
1797 ((uint32_t*)(src+i*stride))[1]= dc1;
|
|
1798 }
|
|
1799 }
|
|
1800
|
|
1801
|
|
1802 static void pred8x8_dc_c(uint8_t *src, int stride){
|
|
1803 int i;
|
|
1804 int dc0, dc1, dc2, dc3;
|
|
1805
|
|
1806 dc0=dc1=dc2=0;
|
|
1807 for(i=0;i<4; i++){
|
|
1808 dc0+= src[-1+i*stride] + src[i-stride];
|
|
1809 dc1+= src[4+i-stride];
|
|
1810 dc2+= src[-1+(i+4)*stride];
|
|
1811 }
|
|
1812 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
|
|
1813 dc0= 0x01010101*((dc0 + 4)>>3);
|
|
1814 dc1= 0x01010101*((dc1 + 2)>>2);
|
|
1815 dc2= 0x01010101*((dc2 + 2)>>2);
|
|
1816
|
|
1817 for(i=0; i<4; i++){
|
|
1818 ((uint32_t*)(src+i*stride))[0]= dc0;
|
|
1819 ((uint32_t*)(src+i*stride))[1]= dc1;
|
|
1820 }
|
|
1821 for(i=4; i<8; i++){
|
|
1822 ((uint32_t*)(src+i*stride))[0]= dc2;
|
|
1823 ((uint32_t*)(src+i*stride))[1]= dc3;
|
|
1824 }
|
|
1825 }
|
|
1826
|
|
1827 static void pred8x8_plane_c(uint8_t *src, int stride){
|
|
1828 uint8_t *cm = cropTbl + MAX_NEG_CROP;
|
|
1829 int i, dx, dy, dc;
|
|
1830 int temp[8];
|
|
1831
|
|
1832 dc= 16*(src[7-stride] + src[-1+7*stride]);
|
|
1833
|
|
1834 dx=dy=0;
|
|
1835 for(i=1; i<5; i++){
|
|
1836 dx += i*(src[3+i-stride] - src[3-i-stride]);
|
|
1837 dy += i*(src[-1+(3+i)*stride] - src[-1+(3-i)*stride]);
|
|
1838 }
|
|
1839 dx= (17*dx+16)>>5;
|
|
1840 dy= (17*dy+16)>>5;
|
|
1841
|
|
1842 dc += 16;
|
|
1843
|
|
1844 //FIXME modifiy dc,dx,dy to avoid -3
|
|
1845
|
|
1846 for(i=0; i<8; i++)
|
|
1847 temp[i]= dx*(i-3) + dc;
|
|
1848
|
|
1849 if( (dc - ABS(dx)*4 - ABS(dy)*4)>>5 < 0
|
|
1850 || (dc + ABS(dx)*4 + ABS(dy)*4)>>5 > 255){
|
|
1851
|
|
1852 for(i=0; i<8; i++){
|
|
1853 int j;
|
|
1854 for(j=0; j<8; j++)
|
|
1855 src[j + i*stride]= cm[ (temp[j] + dy*(i-3))>>5 ];
|
|
1856 }
|
|
1857 }else{
|
|
1858 for(i=0; i<8; i++){
|
|
1859 int j;
|
|
1860 for(j=0; j<8; j++)
|
|
1861 src[j + i*stride]= (temp[j] + dy*(i-3))>>5;
|
|
1862 }
|
|
1863 }
|
|
1864 }
|
|
1865
|
|
1866 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
|
|
1867 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
|
|
1868 int src_x_offset, int src_y_offset,
|
|
1869 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
|
|
1870 MpegEncContext * const s = &h->s;
|
|
1871 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
|
|
1872 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
|
|
1873 const int luma_xy= (mx&3) + ((my&3)<<2);
|
|
1874 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
|
|
1875 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
|
|
1876 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
|
|
1877 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
|
|
1878 int extra_height= extra_width;
|
|
1879 int emu=0;
|
|
1880 const int full_mx= mx>>2;
|
|
1881 const int full_my= my>>2;
|
|
1882
|
|
1883 assert(pic->data[0]);
|
|
1884
|
|
1885 if(mx&7) extra_width -= 3;
|
|
1886 if(my&7) extra_height -= 3;
|
|
1887
|
|
1888 if( full_mx < 0-extra_width
|
|
1889 || full_my < 0-extra_height
|
|
1890 || full_mx + 16/*FIXME*/ > s->width + extra_width
|
|
1891 || full_my + 16/*FIXME*/ > s->height + extra_height){
|
|
1892 ff_emulated_edge_mc(s, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
|
|
1893 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
|
|
1894 emu=1;
|
|
1895 }
|
|
1896
|
|
1897 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
|
|
1898 if(!square){
|
|
1899 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
|
|
1900 }
|
|
1901
|
|
1902 if(s->flags&CODEC_FLAG_GRAY) return;
|
|
1903
|
|
1904 if(emu){
|
|
1905 ff_emulated_edge_mc(s, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
|
|
1906 src_cb= s->edge_emu_buffer;
|
|
1907 }
|
|
1908 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
|
|
1909
|
|
1910 if(emu){
|
|
1911 ff_emulated_edge_mc(s, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
|
|
1912 src_cr= s->edge_emu_buffer;
|
|
1913 }
|
|
1914 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
|
|
1915 }
|
|
1916
|
|
1917 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
|
|
1918 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
|
|
1919 int x_offset, int y_offset,
|
|
1920 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
|
|
1921 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
|
|
1922 int list0, int list1){
|
|
1923 MpegEncContext * const s = &h->s;
|
|
1924 qpel_mc_func *qpix_op= qpix_put;
|
|
1925 h264_chroma_mc_func chroma_op= chroma_put;
|
|
1926
|
|
1927 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
|
|
1928 dest_cb += x_offset + y_offset*s->uvlinesize;
|
|
1929 dest_cr += x_offset + y_offset*s->uvlinesize;
|
|
1930 x_offset += 8*s->mb_x;
|
|
1931 y_offset += 8*s->mb_y;
|
|
1932
|
|
1933 if(list0){
|
1169
|
1934 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
|
1168
|
1935 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
|
|
1936 dest_y, dest_cb, dest_cr, x_offset, y_offset,
|
|
1937 qpix_op, chroma_op);
|
|
1938
|
|
1939 qpix_op= qpix_avg;
|
|
1940 chroma_op= chroma_avg;
|
|
1941 }
|
|
1942
|
|
1943 if(list1){
|
1169
|
1944 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
|
1168
|
1945 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
|
|
1946 dest_y, dest_cb, dest_cr, x_offset, y_offset,
|
|
1947 qpix_op, chroma_op);
|
|
1948 }
|
|
1949 }
|
|
1950
|
|
1951 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
|
|
1952 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
|
|
1953 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
|
|
1954 MpegEncContext * const s = &h->s;
|
|
1955 const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
|
|
1956 const int mb_type= s->current_picture.mb_type[mb_xy];
|
|
1957
|
|
1958 assert(IS_INTER(mb_type));
|
|
1959
|
|
1960 if(IS_16X16(mb_type)){
|
|
1961 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
|
|
1962 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
|
|
1963 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
|
|
1964 }else if(IS_16X8(mb_type)){
|
|
1965 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
|
|
1966 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
|
|
1967 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
|
|
1968 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
|
|
1969 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
|
|
1970 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
|
|
1971 }else if(IS_8X16(mb_type)){
|
|
1972 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
|
|
1973 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
|
|
1974 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
|
|
1975 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
|
|
1976 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
|
|
1977 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
|
|
1978 }else{
|
|
1979 int i;
|
|
1980
|
|
1981 assert(IS_8X8(mb_type));
|
|
1982
|
|
1983 for(i=0; i<4; i++){
|
|
1984 const int sub_mb_type= h->sub_mb_type[i];
|
|
1985 const int n= 4*i;
|
|
1986 int x_offset= (i&1)<<2;
|
|
1987 int y_offset= (i&2)<<1;
|
|
1988
|
|
1989 if(IS_SUB_8X8(sub_mb_type)){
|
|
1990 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
|
|
1991 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
|
|
1992 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
|
|
1993 }else if(IS_SUB_8X4(sub_mb_type)){
|
|
1994 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
|
|
1995 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
|
|
1996 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
|
|
1997 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
|
|
1998 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
|
|
1999 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
|
|
2000 }else if(IS_SUB_4X8(sub_mb_type)){
|
|
2001 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
|
|
2002 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
|
|
2003 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
|
|
2004 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
|
|
2005 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
|
|
2006 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
|
|
2007 }else{
|
|
2008 int j;
|
|
2009 assert(IS_SUB_4X4(sub_mb_type));
|
|
2010 for(j=0; j<4; j++){
|
|
2011 int sub_x_offset= x_offset + 2*(j&1);
|
|
2012 int sub_y_offset= y_offset + (j&2);
|
|
2013 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
|
|
2014 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
|
|
2015 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
|
|
2016 }
|
|
2017 }
|
|
2018 }
|
|
2019 }
|
|
2020 }
|
|
2021
|
|
2022 static void decode_init_vlc(H264Context *h){
|
|
2023 static int done = 0;
|
|
2024
|
|
2025 if (!done) {
|
|
2026 int i;
|
|
2027 done = 1;
|
|
2028
|
|
2029 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
|
|
2030 &chroma_dc_coeff_token_len [0], 1, 1,
|
|
2031 &chroma_dc_coeff_token_bits[0], 1, 1);
|
|
2032
|
|
2033 for(i=0; i<4; i++){
|
|
2034 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
|
|
2035 &coeff_token_len [i][0], 1, 1,
|
|
2036 &coeff_token_bits[i][0], 1, 1);
|
|
2037 }
|
|
2038
|
|
2039 for(i=0; i<3; i++){
|
|
2040 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
|
|
2041 &chroma_dc_total_zeros_len [i][0], 1, 1,
|
|
2042 &chroma_dc_total_zeros_bits[i][0], 1, 1);
|
|
2043 }
|
|
2044 for(i=0; i<15; i++){
|
|
2045 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
|
|
2046 &total_zeros_len [i][0], 1, 1,
|
|
2047 &total_zeros_bits[i][0], 1, 1);
|
|
2048 }
|
|
2049
|
|
2050 for(i=0; i<6; i++){
|
|
2051 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
|
|
2052 &run_len [i][0], 1, 1,
|
|
2053 &run_bits[i][0], 1, 1);
|
|
2054 }
|
|
2055 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
|
|
2056 &run_len [6][0], 1, 1,
|
|
2057 &run_bits[6][0], 1, 1);
|
|
2058 }
|
|
2059 }
|
|
2060
|
|
2061 /**
|
|
2062 * Sets the intra prediction function pointers.
|
|
2063 */
|
|
2064 static void init_pred_ptrs(H264Context *h){
|
|
2065 // MpegEncContext * const s = &h->s;
|
|
2066
|
|
2067 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
|
|
2068 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
|
|
2069 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
|
|
2070 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
|
|
2071 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
|
|
2072 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
|
|
2073 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
|
|
2074 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
|
|
2075 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
|
|
2076 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
|
|
2077 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
|
|
2078 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
|
|
2079
|
|
2080 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
|
|
2081 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
|
|
2082 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
|
|
2083 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
|
|
2084 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
|
|
2085 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
|
|
2086 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
|
|
2087
|
|
2088 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
|
|
2089 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
|
|
2090 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
|
|
2091 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
|
|
2092 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
|
|
2093 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
|
|
2094 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
|
|
2095 }
|
|
2096
|
|
2097 //FIXME factorize
|
|
2098 #define CHECKED_ALLOCZ(p, size)\
|
|
2099 {\
|
|
2100 p= av_mallocz(size);\
|
|
2101 if(p==NULL){\
|
|
2102 perror("malloc");\
|
|
2103 goto fail;\
|
|
2104 }\
|
|
2105 }
|
|
2106
|
|
2107 static void free_tables(H264Context *h){
|
|
2108 MpegEncContext * const s = &h->s;
|
|
2109
|
|
2110 av_freep(&h->intra4x4_pred_mode);
|
|
2111 av_freep(&h->non_zero_count);
|
|
2112 av_freep(&h->slice_table_base);
|
|
2113 h->slice_table= NULL;
|
|
2114
|
|
2115 av_freep(&h->mb2b_xy);
|
|
2116 av_freep(&h->mb2b8_xy);
|
|
2117 }
|
|
2118
|
|
2119 /**
|
|
2120 * allocates tables.
|
|
2121 * needs widzh/height
|
|
2122 */
|
|
2123 static int alloc_tables(H264Context *h){
|
|
2124 MpegEncContext * const s = &h->s;
|
|
2125 const int big_mb_num= h->mb_stride * (s->mb_height+1);
|
|
2126 int x,y;
|
|
2127
|
|
2128 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
|
|
2129 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
|
|
2130 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
|
|
2131
|
|
2132 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
|
|
2133 h->slice_table= h->slice_table_base + h->mb_stride + 1;
|
|
2134
|
|
2135 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
|
|
2136 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
|
|
2137 for(y=0; y<s->mb_height; y++){
|
|
2138 for(x=0; x<s->mb_width; x++){
|
|
2139 const int mb_xy= x + y*h->mb_stride;
|
|
2140 const int b_xy = 4*x + 4*y*h->b_stride;
|
|
2141 const int b8_xy= 2*x + 2*y*h->b8_stride;
|
|
2142
|
|
2143 h->mb2b_xy [mb_xy]= b_xy;
|
|
2144 h->mb2b8_xy[mb_xy]= b8_xy;
|
|
2145 }
|
|
2146 }
|
|
2147
|
|
2148 return 0;
|
|
2149 fail:
|
|
2150 free_tables(h);
|
|
2151 return -1;
|
|
2152 }
|
|
2153
|
|
2154 static void common_init(H264Context *h){
|
|
2155 MpegEncContext * const s = &h->s;
|
|
2156 int i;
|
|
2157
|
|
2158 s->width = s->avctx->width;
|
|
2159 s->height = s->avctx->height;
|
|
2160 s->codec_id= s->avctx->codec->id;
|
|
2161
|
|
2162 init_pred_ptrs(h);
|
|
2163
|
|
2164 s->decode=1; //FIXME
|
|
2165 }
|
|
2166
|
|
2167 static int decode_init(AVCodecContext *avctx){
|
|
2168 H264Context *h= avctx->priv_data;
|
|
2169 MpegEncContext * const s = &h->s;
|
|
2170
|
|
2171 s->avctx = avctx;
|
|
2172 common_init(h);
|
|
2173
|
|
2174 s->out_format = FMT_H264;
|
|
2175 s->workaround_bugs= avctx->workaround_bugs;
|
|
2176
|
|
2177 // set defaults
|
|
2178 s->progressive_sequence=1;
|
|
2179 // s->decode_mb= ff_h263_decode_mb;
|
|
2180 s->low_delay= 1;
|
|
2181 avctx->pix_fmt= PIX_FMT_YUV420P;
|
|
2182
|
|
2183 decode_init_vlc(h);
|
|
2184
|
|
2185 return 0;
|
|
2186 }
|
|
2187
|
|
2188 static void frame_start(H264Context *h){
|
|
2189 MpegEncContext * const s = &h->s;
|
|
2190 int i;
|
|
2191
|
|
2192 MPV_frame_start(s, s->avctx);
|
|
2193 ff_er_frame_start(s);
|
|
2194 h->mmco_index=0;
|
|
2195
|
|
2196 assert(s->linesize && s->uvlinesize);
|
|
2197
|
|
2198 for(i=0; i<16; i++){
|
|
2199 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
|
|
2200 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
|
|
2201 }
|
|
2202 for(i=0; i<4; i++){
|
|
2203 h->block_offset[16+i]=
|
|
2204 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
|
|
2205 }
|
|
2206
|
|
2207 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
|
|
2208 }
|
|
2209
|
|
2210 static void hl_decode_mb(H264Context *h){
|
|
2211 MpegEncContext * const s = &h->s;
|
|
2212 const int mb_x= s->mb_x;
|
|
2213 const int mb_y= s->mb_y;
|
|
2214 const int mb_xy= mb_x + mb_y*h->mb_stride;
|
|
2215 const int mb_type= s->current_picture.mb_type[mb_xy];
|
|
2216 uint8_t *dest_y, *dest_cb, *dest_cr;
|
|
2217 int linesize, uvlinesize /*dct_offset*/;
|
|
2218 int i;
|
|
2219
|
|
2220 if(!s->decode)
|
|
2221 return;
|
|
2222
|
|
2223 if(s->mb_skiped){
|
|
2224 }
|
|
2225
|
|
2226 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
|
|
2227 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
|
|
2228 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
|
|
2229
|
|
2230 if (h->mb_field_decoding_flag) {
|
|
2231 linesize = s->linesize * 2;
|
|
2232 uvlinesize = s->uvlinesize * 2;
|
|
2233 if(mb_y&1){ //FIXME move out of this func?
|
|
2234 dest_y -= s->linesize*15;
|
|
2235 dest_cb-= s->linesize*7;
|
|
2236 dest_cr-= s->linesize*7;
|
|
2237 }
|
|
2238 } else {
|
|
2239 linesize = s->linesize;
|
|
2240 uvlinesize = s->uvlinesize;
|
|
2241 // dct_offset = s->linesize * 16;
|
|
2242 }
|
|
2243
|
|
2244 if(IS_INTRA(mb_type)){
|
|
2245 if(!(s->flags&CODEC_FLAG_GRAY)){
|
|
2246 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
|
|
2247 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
|
|
2248 }
|
|
2249
|
|
2250 if(IS_INTRA4x4(mb_type)){
|
|
2251 if(!s->encoding){
|
|
2252 for(i=0; i<16; i++){
|
|
2253 uint8_t * const ptr= dest_y + h->block_offset[i];
|
|
2254 uint8_t *topright= ptr + 4 - linesize;
|
|
2255 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
|
|
2256 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
|
|
2257 int tr;
|
|
2258
|
|
2259 if(!topright_avail){
|
|
2260 tr= ptr[3 - linesize]*0x01010101;
|
|
2261 topright= (uint8_t*) &tr;
|
|
2262 }
|
|
2263
|
|
2264 h->pred4x4[ dir ](ptr, topright, linesize);
|
|
2265 if(h->non_zero_count_cache[ scan8[i] ])
|
|
2266 h264_add_idct_c(ptr, h->mb + i*16, linesize);
|
|
2267 }
|
|
2268 }
|
|
2269 }else{
|
|
2270 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
|
|
2271 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
|
|
2272 }
|
|
2273 }else{
|
|
2274 hl_motion(h, dest_y, dest_cb, dest_cr,
|
|
2275 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
|
|
2276 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
|
|
2277 }
|
|
2278
|
|
2279
|
|
2280 if(!IS_INTRA4x4(mb_type)){
|
|
2281 for(i=0; i<16; i++){
|
|
2282 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
|
|
2283 uint8_t * const ptr= dest_y + h->block_offset[i];
|
|
2284 h264_add_idct_c(ptr, h->mb + i*16, linesize);
|
|
2285 }
|
|
2286 }
|
|
2287 }
|
|
2288
|
|
2289 if(!(s->flags&CODEC_FLAG_GRAY)){
|
|
2290 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
|
|
2291 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
|
|
2292 for(i=16; i<16+4; i++){
|
|
2293 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
|
|
2294 uint8_t * const ptr= dest_cb + h->block_offset[i];
|
|
2295 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
|
|
2296 }
|
|
2297 }
|
|
2298 for(i=20; i<20+4; i++){
|
|
2299 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
|
|
2300 uint8_t * const ptr= dest_cr + h->block_offset[i];
|
|
2301 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
|
|
2302 }
|
|
2303 }
|
|
2304 }
|
|
2305 }
|
|
2306
|
|
2307 static void decode_mb_cabac(H264Context *h){
|
|
2308 // MpegEncContext * const s = &h->s;
|
|
2309 }
|
|
2310
|
|
2311 /**
|
|
2312 * fills the default_ref_list.
|
|
2313 */
|
|
2314 static int fill_default_ref_list(H264Context *h){
|
|
2315 MpegEncContext * const s = &h->s;
|
|
2316 int i;
|
|
2317 Picture sorted_short_ref[16];
|
|
2318
|
|
2319 if(h->slice_type==B_TYPE){
|
|
2320 int out_i;
|
|
2321 int limit= -1;
|
|
2322
|
|
2323 for(out_i=0; out_i<h->short_ref_count; out_i++){
|
|
2324 int best_i=-1;
|
|
2325 int best_poc=-1;
|
|
2326
|
|
2327 for(i=0; i<h->short_ref_count; i++){
|
|
2328 const int poc= h->short_ref[i]->poc;
|
|
2329 if(poc > limit && poc < best_poc){
|
|
2330 best_poc= poc;
|
|
2331 best_i= i;
|
|
2332 }
|
|
2333 }
|
|
2334
|
|
2335 assert(best_i != -1);
|
|
2336
|
|
2337 limit= best_poc;
|
|
2338 sorted_short_ref[out_i]= *h->short_ref[best_i];
|
|
2339 }
|
|
2340 }
|
|
2341
|
|
2342 if(s->picture_structure == PICT_FRAME){
|
|
2343 if(h->slice_type==B_TYPE){
|
|
2344 const int current_poc= s->current_picture_ptr->poc;
|
|
2345 int list;
|
|
2346
|
|
2347 for(list=0; list<2; list++){
|
|
2348 int index=0;
|
|
2349
|
|
2350 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
|
|
2351 const int i2= list ? h->short_ref_count - i - 1 : i;
|
|
2352 const int poc= sorted_short_ref[i2].poc;
|
|
2353
|
|
2354 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
|
|
2355
|
|
2356 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
|
|
2357 h->default_ref_list[list][index ]= sorted_short_ref[i2];
|
|
2358 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
|
|
2359 }
|
|
2360 }
|
|
2361
|
|
2362 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
|
|
2363 if(h->long_ref[i]->reference != 3) continue;
|
|
2364
|
|
2365 h->default_ref_list[ list ][index ]= *h->long_ref[i];
|
|
2366 h->default_ref_list[ list ][index++].pic_id= i;;
|
|
2367 }
|
|
2368
|
|
2369 if(h->long_ref_count > 1 && h->short_ref_count==0){
|
|
2370 Picture temp= h->default_ref_list[1][0];
|
|
2371 h->default_ref_list[1][0] = h->default_ref_list[1][1];
|
|
2372 h->default_ref_list[1][0] = temp;
|
|
2373 }
|
|
2374
|
|
2375 if(index < h->ref_count[ list ])
|
|
2376 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
|
|
2377 }
|
|
2378 }else{
|
|
2379 int index=0;
|
|
2380 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
|
|
2381 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
|
|
2382 h->default_ref_list[0][index ]= *h->short_ref[i];
|
|
2383 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
|
|
2384 }
|
|
2385 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
|
|
2386 if(h->long_ref[i]->reference != 3) continue;
|
|
2387 h->default_ref_list[0][index ]= *h->long_ref[i];
|
|
2388 h->default_ref_list[0][index++].pic_id= i;;
|
|
2389 }
|
|
2390 if(index < h->ref_count[0])
|
|
2391 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
|
|
2392 }
|
|
2393 }else{ //FIELD
|
|
2394 if(h->slice_type==B_TYPE){
|
|
2395 }else{
|
|
2396 //FIXME second field balh
|
|
2397 }
|
|
2398 }
|
|
2399 return 0;
|
|
2400 }
|
|
2401
|
|
2402 static int decode_ref_pic_list_reordering(H264Context *h){
|
|
2403 MpegEncContext * const s = &h->s;
|
|
2404 int list;
|
|
2405
|
|
2406 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
|
|
2407
|
|
2408 for(list=0; list<2; list++){
|
|
2409 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
|
|
2410
|
|
2411 if(get_bits1(&s->gb)){
|
|
2412 int pred= h->curr_pic_num;
|
|
2413 int index;
|
|
2414
|
|
2415 for(index=0; ; index++){
|
|
2416 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
|
|
2417 int pic_id;
|
|
2418 int i;
|
|
2419
|
|
2420
|
|
2421 if(index >= h->ref_count[list]){
|
|
2422 fprintf(stderr, "reference count overflow\n");
|
|
2423 return -1;
|
|
2424 }
|
|
2425
|
|
2426 if(reordering_of_pic_nums_idc<3){
|
|
2427 if(reordering_of_pic_nums_idc<2){
|
|
2428 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
|
|
2429
|
|
2430 if(abs_diff_pic_num >= h->max_pic_num){
|
|
2431 fprintf(stderr, "abs_diff_pic_num overflow\n");
|
|
2432 return -1;
|
|
2433 }
|
|
2434
|
|
2435 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
|
|
2436 else pred+= abs_diff_pic_num;
|
|
2437 pred &= h->max_pic_num - 1;
|
|
2438
|
|
2439 for(i= h->ref_count[list]-1; i>=index; i--){
|
|
2440 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
|
|
2441 break;
|
|
2442 }
|
|
2443 }else{
|
|
2444 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
|
|
2445
|
|
2446 for(i= h->ref_count[list]-1; i>=index; i--){
|
|
2447 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
|
|
2448 break;
|
|
2449 }
|
|
2450 }
|
|
2451
|
|
2452 if(i < index){
|
|
2453 fprintf(stderr, "reference picture missing during reorder\n");
|
|
2454 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
|
|
2455 }else if(i > index){
|
|
2456 Picture tmp= h->ref_list[list][i];
|
|
2457 for(; i>index; i--){
|
|
2458 h->ref_list[list][i]= h->ref_list[list][i-1];
|
|
2459 }
|
|
2460 h->ref_list[list][index]= tmp;
|
|
2461 }
|
|
2462 }else if(reordering_of_pic_nums_idc==3)
|
|
2463 break;
|
|
2464 else{
|
|
2465 fprintf(stderr, "illegal reordering_of_pic_nums_idc\n");
|
|
2466 return -1;
|
|
2467 }
|
|
2468 }
|
|
2469 }
|
|
2470
|
|
2471 if(h->slice_type!=B_TYPE) break;
|
|
2472 }
|
|
2473 return 0;
|
|
2474 }
|
|
2475
|
|
2476 static int pred_weight_table(H264Context *h){
|
|
2477 MpegEncContext * const s = &h->s;
|
|
2478 int list, i;
|
|
2479
|
|
2480 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
|
|
2481 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
|
|
2482
|
|
2483 for(list=0; list<2; list++){
|
|
2484 for(i=0; i<h->ref_count[list]; i++){
|
|
2485 int luma_weight_flag, chroma_weight_flag;
|
|
2486
|
|
2487 luma_weight_flag= get_bits1(&s->gb);
|
|
2488 if(luma_weight_flag){
|
|
2489 h->luma_weight[list][i]= get_se_golomb(&s->gb);
|
|
2490 h->luma_offset[list][i]= get_se_golomb(&s->gb);
|
|
2491 }
|
|
2492
|
|
2493 chroma_weight_flag= get_bits1(&s->gb);
|
|
2494 if(chroma_weight_flag){
|
|
2495 int j;
|
|
2496 for(j=0; j<2; j++){
|
|
2497 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
|
|
2498 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
|
|
2499 }
|
|
2500 }
|
|
2501 }
|
|
2502 if(h->slice_type != B_TYPE) break;
|
|
2503 }
|
|
2504 return 0;
|
|
2505 }
|
|
2506
|
|
2507 /**
|
|
2508 * instantaneos decoder refresh.
|
|
2509 */
|
|
2510 static void idr(H264Context *h){
|
|
2511 int i;
|
|
2512
|
|
2513 for(i=0; i<h->long_ref_count; i++){
|
|
2514 h->long_ref[i]->reference=0;
|
|
2515 h->long_ref[i]= NULL;
|
|
2516 }
|
|
2517 h->long_ref_count=0;
|
|
2518
|
|
2519 for(i=0; i<h->short_ref_count; i++){
|
|
2520 h->short_ref[i]->reference=0;
|
|
2521 h->short_ref[i]= NULL;
|
|
2522 }
|
|
2523 h->short_ref_count=0;
|
|
2524 }
|
|
2525
|
|
2526 /**
|
|
2527 *
|
|
2528 * @return the removed picture or NULL if an error occures
|
|
2529 */
|
|
2530 static Picture * remove_short(H264Context *h, int frame_num){
|
1169
|
2531 MpegEncContext * const s = &h->s;
|
1168
|
2532 int i;
|
|
2533
|
1169
|
2534 if(s->avctx->debug&FF_DEBUG_MMCO)
|
|
2535 printf("remove short %d count %d\n", frame_num, h->short_ref_count);
|
|
2536
|
1168
|
2537 for(i=0; i<h->short_ref_count; i++){
|
|
2538 Picture *pic= h->short_ref[i];
|
1169
|
2539 if(s->avctx->debug&FF_DEBUG_MMCO)
|
|
2540 printf("%d %d %X\n", i, pic->frame_num, (int)pic);
|
1168
|
2541 if(pic->frame_num == frame_num){
|
|
2542 h->short_ref[i]= NULL;
|
|
2543 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
|
|
2544 h->short_ref_count--;
|
|
2545 return pic;
|
|
2546 }
|
|
2547 }
|
|
2548 return NULL;
|
|
2549 }
|
|
2550
|
|
2551 /**
|
|
2552 *
|
|
2553 * @return the removed picture or NULL if an error occures
|
|
2554 */
|
|
2555 static Picture * remove_long(H264Context *h, int i){
|
|
2556 Picture *pic;
|
|
2557
|
|
2558 if(i >= h->long_ref_count) return NULL;
|
|
2559 pic= h->long_ref[i];
|
|
2560 if(pic==NULL) return NULL;
|
|
2561
|
|
2562 h->long_ref[i]= NULL;
|
|
2563 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
|
|
2564 h->long_ref_count--;
|
|
2565
|
|
2566 return pic;
|
|
2567 }
|
|
2568
|
|
2569 /**
|
|
2570 * Executes the reference picture marking (memory management control operations).
|
|
2571 */
|
|
2572 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
|
|
2573 MpegEncContext * const s = &h->s;
|
|
2574 int i;
|
|
2575 int current_is_long=0;
|
|
2576 Picture *pic;
|
|
2577
|
|
2578 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
|
|
2579 printf("no mmco here\n");
|
|
2580
|
|
2581 for(i=0; i<mmco_count; i++){
|
|
2582 if(s->avctx->debug&FF_DEBUG_MMCO)
|
|
2583 printf("mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
|
|
2584
|
|
2585 switch(mmco[i].opcode){
|
|
2586 case MMCO_SHORT2UNUSED:
|
|
2587 pic= remove_short(h, mmco[i].short_frame_num);
|
|
2588 if(pic==NULL) return -1;
|
|
2589 pic->reference= 0;
|
|
2590 break;
|
|
2591 case MMCO_SHORT2LONG:
|
|
2592 pic= remove_long(h, mmco[i].long_index);
|
|
2593 if(pic) pic->reference=0;
|
|
2594
|
|
2595 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
|
|
2596 h->long_ref[ mmco[i].long_index ]->long_ref=1;
|
|
2597 break;
|
|
2598 case MMCO_LONG2UNUSED:
|
|
2599 pic= remove_long(h, mmco[i].long_index);
|
|
2600 if(pic==NULL) return -1;
|
|
2601 pic->reference= 0;
|
|
2602 break;
|
|
2603 case MMCO_LONG:
|
|
2604 pic= remove_long(h, mmco[i].long_index);
|
|
2605 if(pic) pic->reference=0;
|
|
2606
|
|
2607 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
|
|
2608 h->long_ref[ mmco[i].long_index ]->long_ref=1;
|
|
2609 h->long_ref_count++;
|
|
2610
|
|
2611 current_is_long=1;
|
|
2612 break;
|
|
2613 case MMCO_SET_MAX_LONG:
|
|
2614 assert(mmco[i].long_index <= 16);
|
|
2615 while(mmco[i].long_index < h->long_ref_count){
|
|
2616 pic= remove_long(h, mmco[i].long_index);
|
|
2617 pic->reference=0;
|
|
2618 }
|
|
2619 while(mmco[i].long_index > h->long_ref_count){
|
|
2620 h->long_ref[ h->long_ref_count++ ]= NULL;
|
|
2621 }
|
|
2622 break;
|
|
2623 case MMCO_RESET:
|
|
2624 while(h->short_ref_count){
|
|
2625 pic= remove_short(h, h->short_ref[0]->frame_num);
|
|
2626 pic->reference=0;
|
|
2627 }
|
|
2628 while(h->long_ref_count){
|
|
2629 pic= remove_long(h, h->long_ref_count-1);
|
|
2630 pic->reference=0;
|
|
2631 }
|
|
2632 break;
|
|
2633 default: assert(0);
|
|
2634 }
|
|
2635 }
|
|
2636
|
|
2637 if(!current_is_long){
|
|
2638 pic= remove_short(h, s->current_picture_ptr->frame_num);
|
|
2639 if(pic){
|
|
2640 pic->reference=0;
|
|
2641 fprintf(stderr, "illegal short term buffer state detected\n");
|
|
2642 }
|
|
2643
|
|
2644 if(h->short_ref_count)
|
1169
|
2645 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
|
|
2646
|
|
2647 h->short_ref[0]= s->current_picture_ptr;
|
1168
|
2648 h->short_ref[0]->long_ref=0;
|
|
2649 h->short_ref_count++;
|
|
2650 }
|
|
2651
|
|
2652 return 0;
|
|
2653 }
|
|
2654
|
|
2655 static int decode_ref_pic_marking(H264Context *h){
|
|
2656 MpegEncContext * const s = &h->s;
|
|
2657 int i;
|
|
2658
|
|
2659 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
|
|
2660 s->broken_link= get_bits1(&s->gb) -1;
|
|
2661 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
|
|
2662 if(h->mmco[0].long_index == -1)
|
|
2663 h->mmco_index= 0;
|
|
2664 else{
|
|
2665 h->mmco[0].opcode= MMCO_LONG;
|
|
2666 h->mmco_index= 1;
|
|
2667 }
|
|
2668 }else{
|
|
2669 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
|
|
2670 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) {
|
|
2671 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
|
|
2672
|
|
2673 h->mmco[i].opcode= opcode;
|
|
2674 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
|
|
2675 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
|
|
2676 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
|
|
2677 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
|
|
2678 return -1;
|
|
2679 }*/
|
|
2680 }
|
|
2681 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
|
|
2682 h->mmco[i].long_index= get_ue_golomb(&s->gb);
|
|
2683 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){
|
|
2684 fprintf(stderr, "illegal long ref in memory management control operation %d\n", opcode);
|
|
2685 return -1;
|
|
2686 }
|
|
2687 }
|
|
2688
|
|
2689 if(opcode > MMCO_LONG){
|
|
2690 fprintf(stderr, "illegal memory management control operation %d\n", opcode);
|
|
2691 return -1;
|
|
2692 }
|
|
2693 }
|
|
2694 h->mmco_index= i;
|
|
2695 }else{
|
|
2696 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
|
|
2697
|
|
2698 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
|
|
2699 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
|
|
2700 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
|
|
2701 h->mmco_index= 1;
|
|
2702 }else
|
|
2703 h->mmco_index= 0;
|
|
2704 }
|
|
2705 }
|
|
2706
|
|
2707 return 0;
|
|
2708 }
|
|
2709
|
|
2710 static int init_poc(H264Context *h){
|
|
2711 MpegEncContext * const s = &h->s;
|
|
2712 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
|
|
2713 int field_poc[2];
|
|
2714
|
|
2715 if(h->nal_unit_type == NAL_IDR_SLICE){
|
|
2716 h->frame_num_offset= 0;
|
|
2717 }else{
|
|
2718 if(h->frame_num < h->prev_frame_num)
|
|
2719 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
|
|
2720 else
|
|
2721 h->frame_num_offset= h->prev_frame_num_offset;
|
|
2722 }
|
|
2723
|
|
2724 if(h->sps.poc_type==0){
|
|
2725 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
|
|
2726
|
|
2727 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
|
|
2728 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
|
|
2729 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
|
|
2730 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
|
|
2731 else
|
|
2732 h->poc_msb = h->prev_poc_msb;
|
|
2733 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
|
|
2734 field_poc[0] =
|
|
2735 field_poc[1] = h->poc_msb + h->poc_lsb;
|
|
2736 if(s->picture_structure == PICT_FRAME)
|
|
2737 field_poc[1] += h->delta_poc_bottom;
|
|
2738 }else if(h->sps.poc_type==1){
|
|
2739 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
|
|
2740 int i;
|
|
2741
|
|
2742 if(h->sps.poc_cycle_length != 0)
|
|
2743 abs_frame_num = h->frame_num_offset + h->frame_num;
|
|
2744 else
|
|
2745 abs_frame_num = 0;
|
|
2746
|
|
2747 if(h->nal_ref_idc==0 && abs_frame_num > 0)
|
|
2748 abs_frame_num--;
|
|
2749
|
|
2750 expected_delta_per_poc_cycle = 0;
|
|
2751 for(i=0; i < h->sps.poc_cycle_length; i++)
|
|
2752 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
|
|
2753
|
|
2754 if(abs_frame_num > 0){
|
|
2755 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
|
|
2756 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
|
|
2757
|
|
2758 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
|
|
2759 for(i = 0; i <= frame_num_in_poc_cycle; i++)
|
|
2760 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
|
|
2761 } else
|
|
2762 expectedpoc = 0;
|
|
2763
|
|
2764 if(h->nal_ref_idc == 0)
|
|
2765 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
|
|
2766
|
|
2767 field_poc[0] = expectedpoc + h->delta_poc[0];
|
|
2768 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
|
|
2769
|
|
2770 if(s->picture_structure == PICT_FRAME)
|
|
2771 field_poc[1] += h->delta_poc[1];
|
|
2772 }else{
|
|
2773 int poc;
|
|
2774 if(h->nal_unit_type == NAL_IDR_SLICE){
|
|
2775 poc= 0;
|
|
2776 }else{
|
|
2777 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
|
|
2778 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
|
|
2779 }
|
|
2780 field_poc[0]= poc;
|
|
2781 field_poc[1]= poc;
|
|
2782 }
|
|
2783
|
|
2784 if(s->picture_structure != PICT_BOTTOM_FIELD)
|
|
2785 s->current_picture_ptr->field_poc[0]= field_poc[0];
|
|
2786 if(s->picture_structure != PICT_TOP_FIELD)
|
|
2787 s->current_picture_ptr->field_poc[1]= field_poc[1];
|
|
2788 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
|
|
2789 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
|
|
2790
|
|
2791 return 0;
|
|
2792 }
|
|
2793
|
|
2794 /**
|
|
2795 * decodes a slice header.
|
|
2796 * this will allso call MPV_common_init() and frame_start() as needed
|
|
2797 */
|
|
2798 static int decode_slice_header(H264Context *h){
|
|
2799 MpegEncContext * const s = &h->s;
|
|
2800 int first_mb_in_slice, pps_id;
|
|
2801 int num_ref_idx_active_override_flag;
|
|
2802 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
|
|
2803 float new_aspect;
|
|
2804
|
|
2805 s->current_picture.reference= h->nal_ref_idc != 0;
|
|
2806
|
|
2807 first_mb_in_slice= get_ue_golomb(&s->gb);
|
|
2808
|
|
2809 h->slice_type= get_ue_golomb(&s->gb);
|
|
2810 if(h->slice_type > 9){
|
|
2811 fprintf(stderr, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
|
|
2812 }
|
|
2813 if(h->slice_type > 4){
|
|
2814 h->slice_type -= 5;
|
|
2815 h->slice_type_fixed=1;
|
|
2816 }else
|
|
2817 h->slice_type_fixed=0;
|
|
2818
|
|
2819 h->slice_type= slice_type_map[ h->slice_type ];
|
|
2820
|
|
2821 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
|
|
2822
|
|
2823 pps_id= get_ue_golomb(&s->gb);
|
|
2824 if(pps_id>255){
|
|
2825 fprintf(stderr, "pps_id out of range\n");
|
|
2826 return -1;
|
|
2827 }
|
|
2828 h->pps= h->pps_buffer[pps_id];
|
1174
|
2829 if(h->pps.slice_group_count == 0){
|
|
2830 fprintf(stderr, "non existing PPS referenced\n");
|
|
2831 return -1;
|
|
2832 }
|
|
2833
|
1168
|
2834 h->sps= h->sps_buffer[ h->pps.sps_id ];
|
1174
|
2835 if(h->sps.log2_max_frame_num == 0){
|
|
2836 fprintf(stderr, "non existing SPS referenced\n");
|
|
2837 return -1;
|
|
2838 }
|
1168
|
2839
|
|
2840 s->mb_width= h->sps.mb_width;
|
|
2841 s->mb_height= h->sps.mb_height;
|
|
2842 h->mb_stride= s->mb_width + 1;
|
|
2843
|
|
2844 h->b_stride= s->mb_width*4;
|
|
2845 h->b8_stride= s->mb_width*2;
|
|
2846
|
|
2847 s->mb_x = first_mb_in_slice % s->mb_width;
|
|
2848 s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
|
|
2849
|
|
2850 s->width = 16*s->mb_width - 2*(h->pps.crop_left + h->pps.crop_right );
|
|
2851 if(h->sps.frame_mbs_only_flag)
|
|
2852 s->height= 16*s->mb_height - 2*(h->pps.crop_top + h->pps.crop_bottom);
|
|
2853 else
|
|
2854 s->height= 16*s->mb_height - 4*(h->pps.crop_top + h->pps.crop_bottom); //FIXME recheck
|
|
2855
|
|
2856 if(h->pps.crop_left || h->pps.crop_top){
|
|
2857 fprintf(stderr, "insane croping not completly supported, this could look slightly wrong ...\n");
|
|
2858 }
|
|
2859
|
|
2860 if(s->aspected_height) //FIXME emms at end of slice ?
|
|
2861 new_aspect= h->sps.sar_width*s->width / (float)(s->height*h->sps.sar_height);
|
|
2862 else
|
|
2863 new_aspect=0;
|
|
2864
|
|
2865 if (s->context_initialized
|
|
2866 && ( s->width != s->avctx->width || s->height != s->avctx->height
|
|
2867 || ABS(new_aspect - s->avctx->aspect_ratio) > 0.001)) {
|
|
2868 free_tables(h);
|
|
2869 MPV_common_end(s);
|
|
2870 }
|
|
2871 if (!s->context_initialized) {
|
|
2872 if (MPV_common_init(s) < 0)
|
|
2873 return -1;
|
|
2874
|
|
2875 alloc_tables(h);
|
|
2876
|
|
2877 s->avctx->width = s->width;
|
|
2878 s->avctx->height = s->height;
|
|
2879 s->avctx->aspect_ratio= new_aspect;
|
|
2880 }
|
|
2881
|
|
2882 if(first_mb_in_slice == 0){
|
|
2883 frame_start(h);
|
|
2884 }
|
|
2885
|
1169
|
2886 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
|
1168
|
2887 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
|
|
2888
|
|
2889 if(h->sps.frame_mbs_only_flag){
|
|
2890 s->picture_structure= PICT_FRAME;
|
|
2891 }else{
|
|
2892 if(get_bits1(&s->gb)) //field_pic_flag
|
|
2893 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
|
|
2894 else
|
|
2895 s->picture_structure= PICT_FRAME;
|
|
2896 }
|
|
2897
|
|
2898 if(s->picture_structure==PICT_FRAME){
|
|
2899 h->curr_pic_num= h->frame_num;
|
|
2900 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
|
|
2901 }else{
|
|
2902 h->curr_pic_num= 2*h->frame_num;
|
|
2903 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
|
|
2904 }
|
|
2905
|
|
2906 if(h->nal_unit_type == NAL_IDR_SLICE){
|
|
2907 int idr_pic_id= get_ue_golomb(&s->gb);
|
|
2908 }
|
|
2909
|
|
2910 if(h->sps.poc_type==0){
|
|
2911 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
|
|
2912
|
|
2913 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
|
|
2914 h->delta_poc_bottom= get_se_golomb(&s->gb);
|
|
2915 }
|
|
2916 }
|
|
2917
|
|
2918 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
|
|
2919 h->delta_poc[0]= get_se_golomb(&s->gb);
|
|
2920
|
|
2921 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
|
|
2922 h->delta_poc[1]= get_se_golomb(&s->gb);
|
|
2923 }
|
|
2924
|
|
2925 init_poc(h);
|
|
2926
|
|
2927 if(h->pps.redundant_pic_cnt_present){
|
|
2928 h->redundant_pic_count= get_ue_golomb(&s->gb);
|
|
2929 }
|
|
2930
|
|
2931 //set defaults, might be overriden a few line later
|
|
2932 h->ref_count[0]= h->pps.ref_count[0];
|
|
2933 h->ref_count[1]= h->pps.ref_count[1];
|
|
2934
|
|
2935 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
|
|
2936 if(h->slice_type == B_TYPE){
|
|
2937 h->direct_spatial_mv_pred= get_bits1(&s->gb);
|
|
2938 }
|
|
2939 num_ref_idx_active_override_flag= get_bits1(&s->gb);
|
|
2940
|
|
2941 if(num_ref_idx_active_override_flag){
|
|
2942 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
|
|
2943 if(h->slice_type==B_TYPE)
|
|
2944 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
|
|
2945
|
|
2946 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
|
|
2947 fprintf(stderr, "reference overflow\n");
|
|
2948 return -1;
|
|
2949 }
|
|
2950 }
|
|
2951 }
|
|
2952
|
|
2953 if(first_mb_in_slice == 0){
|
|
2954 fill_default_ref_list(h);
|
|
2955 }
|
|
2956
|
|
2957 decode_ref_pic_list_reordering(h);
|
|
2958
|
|
2959 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
|
|
2960 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
|
|
2961 pred_weight_table(h);
|
|
2962
|
|
2963 if(s->current_picture.reference)
|
|
2964 decode_ref_pic_marking(h);
|
|
2965 //FIXME CABAC stuff
|
|
2966
|
|
2967 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta
|
|
2968 //FIXME qscale / qp ... stuff
|
|
2969 if(h->slice_type == SP_TYPE){
|
|
2970 int sp_for_switch_flag= get_bits1(&s->gb);
|
|
2971 }
|
|
2972 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
|
|
2973 int slice_qs_delta= get_se_golomb(&s->gb);
|
|
2974 }
|
|
2975
|
|
2976 if( h->pps.deblocking_filter_parameters_present ) {
|
|
2977 h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb);
|
|
2978 if( h->disable_deblocking_filter_idc != 1 ) {
|
|
2979 h->slice_alpha_c0_offset_div2= get_se_golomb(&s->gb);
|
|
2980 h->slice_beta_offset_div2= get_se_golomb(&s->gb);
|
|
2981 }
|
|
2982 }else
|
|
2983 h->disable_deblocking_filter_idc= 0;
|
|
2984
|
|
2985 #if 0 //FMO
|
|
2986 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
|
|
2987 slice_group_change_cycle= get_bits(&s->gb, ?);
|
|
2988 #endif
|
|
2989
|
|
2990 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
|
|
2991 printf("mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n",
|
|
2992 first_mb_in_slice,
|
|
2993 ff_get_pict_type_char(h->slice_type),
|
|
2994 pps_id, h->frame_num,
|
|
2995 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
|
|
2996 h->ref_count[0], h->ref_count[1],
|
|
2997 s->qscale,
|
|
2998 h->disable_deblocking_filter_idc
|
|
2999 );
|
|
3000 }
|
|
3001
|
|
3002 return 0;
|
|
3003 }
|
|
3004
|
|
3005 /**
|
|
3006 *
|
|
3007 */
|
|
3008 static inline int get_level_prefix(GetBitContext *gb){
|
|
3009 unsigned int buf;
|
|
3010 int log;
|
|
3011
|
|
3012 OPEN_READER(re, gb);
|
|
3013 UPDATE_CACHE(re, gb);
|
|
3014 buf=GET_CACHE(re, gb);
|
|
3015
|
|
3016 log= 32 - av_log2(buf);
|
|
3017 #ifdef TRACE
|
|
3018 print_bin(buf>>(32-log), log);
|
|
3019 printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
|
|
3020 #endif
|
|
3021
|
|
3022 LAST_SKIP_BITS(re, gb, log);
|
|
3023 CLOSE_READER(re, gb);
|
|
3024
|
|
3025 return log-1;
|
|
3026 }
|
|
3027
|
|
3028 /**
|
|
3029 * decodes a residual block.
|
|
3030 * @param n block index
|
|
3031 * @param scantable scantable
|
|
3032 * @param max_coeff number of coefficients in the block
|
|
3033 * @return <0 if an error occured
|
|
3034 */
|
|
3035 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
|
|
3036 MpegEncContext * const s = &h->s;
|
|
3037 const uint16_t *qmul= dequant_coeff[qp];
|
|
3038 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};
|
|
3039 int level[16], run[16];
|
|
3040 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
|
|
3041
|
|
3042 //FIXME put trailing_onex into the context
|
|
3043
|
|
3044 if(n == CHROMA_DC_BLOCK_INDEX){
|
|
3045 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
|
|
3046 total_coeff= coeff_token>>2;
|
|
3047 }else{
|
|
3048 if(n == LUMA_DC_BLOCK_INDEX){
|
|
3049 total_coeff= pred_non_zero_count(h, 0);
|
|
3050 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
|
|
3051 total_coeff= coeff_token>>2;
|
|
3052 }else{
|
|
3053 total_coeff= pred_non_zero_count(h, n);
|
|
3054 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
|
|
3055 total_coeff= coeff_token>>2;
|
|
3056 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
|
|
3057 }
|
|
3058 }
|
|
3059
|
|
3060 //FIXME set last_non_zero?
|
|
3061
|
|
3062 if(total_coeff==0)
|
|
3063 return 0;
|
|
3064
|
|
3065 trailing_ones= coeff_token&3;
|
1170
|
3066 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
|
1168
|
3067 assert(total_coeff<=16);
|
|
3068
|
|
3069 for(i=0; i<trailing_ones; i++){
|
|
3070 level[i]= 1 - 2*get_bits1(gb);
|
|
3071 }
|
|
3072
|
|
3073 suffix_length= total_coeff > 10 && trailing_ones < 3;
|
|
3074
|
|
3075 for(; i<total_coeff; i++){
|
|
3076 const int prefix= get_level_prefix(gb);
|
|
3077 int level_code, mask;
|
|
3078
|
|
3079 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
|
|
3080 if(suffix_length)
|
|
3081 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
|
|
3082 else
|
|
3083 level_code= (prefix<<suffix_length); //part
|
|
3084 }else if(prefix==14){
|
|
3085 if(suffix_length)
|
|
3086 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
|
|
3087 else
|
|
3088 level_code= prefix + get_bits(gb, 4); //part
|
|
3089 }else if(prefix==15){
|
|
3090 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
|
|
3091 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
|
|
3092 }else{
|
|
3093 fprintf(stderr, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
|
|
3094 return -1;
|
|
3095 }
|
|
3096
|
|
3097 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
|
|
3098
|
|
3099 mask= -(level_code&1);
|
|
3100 level[i]= (((2+level_code)>>1) ^ mask) - mask;
|
|
3101
|
|
3102 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
|
|
3103
|
|
3104 #if 1
|
|
3105 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
|
|
3106 #else
|
|
3107 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
|
|
3108 ? == prefix > 2 or sth
|
|
3109 #endif
|
1170
|
3110 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
|
1168
|
3111 }
|
|
3112
|
|
3113 if(total_coeff == max_coeff)
|
|
3114 zeros_left=0;
|
|
3115 else{
|
|
3116 if(n == CHROMA_DC_BLOCK_INDEX)
|
|
3117 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
|
|
3118 else
|
|
3119 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
|
|
3120 }
|
|
3121
|
|
3122 for(i=0; i<total_coeff-1; i++){
|
|
3123 if(zeros_left <=0)
|
|
3124 break;
|
|
3125 else if(zeros_left < 7){
|
|
3126 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
|
|
3127 }else{
|
|
3128 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
|
|
3129 }
|
|
3130 zeros_left -= run[i];
|
|
3131 }
|
|
3132
|
|
3133 if(zeros_left<0){
|
|
3134 fprintf(stderr, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
|
|
3135 return -1;
|
|
3136 }
|
|
3137
|
|
3138 for(; i<total_coeff-1; i++){
|
|
3139 run[i]= 0;
|
|
3140 }
|
|
3141
|
|
3142 run[i]= zeros_left;
|
|
3143
|
|
3144 coeff_num=-1;
|
|
3145 if(n > 24){
|
|
3146 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
|
|
3147 int j;
|
|
3148
|
|
3149 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
|
|
3150 j= scantable[ coeff_num ];
|
|
3151
|
|
3152 block[j]= level[i];
|
|
3153 }
|
|
3154 }else{
|
|
3155 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
|
|
3156 int j;
|
|
3157
|
|
3158 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
|
|
3159 j= scantable[ coeff_num ];
|
|
3160
|
|
3161 block[j]= level[i] * qmul[j];
|
|
3162 // printf("%d %d ", block[j], qmul[j]);
|
|
3163 }
|
|
3164 }
|
|
3165 return 0;
|
|
3166 }
|
|
3167
|
|
3168 /**
|
|
3169 * decodes a macroblock
|
|
3170 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
|
|
3171 */
|
|
3172 static int decode_mb(H264Context *h){
|
|
3173 MpegEncContext * const s = &h->s;
|
|
3174 const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
|
1169
|
3175 int mb_type, partition_count, cbp;
|
1168
|
3176
|
|
3177 memset(h->mb, 0, sizeof(int16_t)*24*16); //FIXME avoid if allready clear (move after skip handlong?
|
|
3178
|
1170
|
3179 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
|
1168
|
3180
|
|
3181 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
|
|
3182 if(s->mb_skip_run==-1)
|
|
3183 s->mb_skip_run= get_ue_golomb(&s->gb);
|
|
3184
|
|
3185 if (s->mb_skip_run--) {
|
|
3186 int i, mx, my;
|
|
3187 /* skip mb */
|
|
3188 #if 0 //FIXME
|
|
3189 for(i=0;i<6;i++)
|
|
3190 s->block_last_index[i] = -1;
|
|
3191 s->mv_type = MV_TYPE_16X16;
|
|
3192 /* if P type, zero motion vector is implied */
|
|
3193 s->mv_dir = MV_DIR_FORWARD;
|
|
3194 s->mb_skiped = 1;
|
|
3195 #endif
|
|
3196 //FIXME b frame
|
|
3197 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
|
|
3198
|
|
3199 memset(h->non_zero_count[mb_xy], 0, 16);
|
|
3200 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
|
|
3201
|
|
3202 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
|
|
3203 h->mb_field_decoding_flag= get_bits1(&s->gb);
|
|
3204 }
|
|
3205
|
|
3206 if(h->mb_field_decoding_flag)
|
|
3207 mb_type|= MB_TYPE_INTERLACED;
|
|
3208
|
|
3209 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
|
|
3210 pred_pskip_motion(h, &mx, &my);
|
|
3211 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
|
|
3212 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, (mx&0xFFFF)+(my<<16), 4);
|
|
3213 write_back_motion(h, mb_type);
|
|
3214
|
|
3215 s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
|
|
3216 h->slice_table[ mb_xy ]= h->slice_num;
|
|
3217
|
|
3218 h->prev_mb_skiped= 1;
|
|
3219 return 0;
|
|
3220 }
|
|
3221 }
|
|
3222 if(h->sps.mb_aff /* && !field pic FIXME needed? */){
|
|
3223 if((s->mb_y&1)==0)
|
|
3224 h->mb_field_decoding_flag = get_bits1(&s->gb);
|
|
3225 }else
|
|
3226 h->mb_field_decoding_flag=0; //FIXME som ed note ?!
|
|
3227
|
|
3228 h->prev_mb_skiped= 0;
|
|
3229
|
|
3230 mb_type= get_ue_golomb(&s->gb);
|
|
3231 if(h->slice_type == B_TYPE){
|
|
3232 if(mb_type < 23){
|
|
3233 partition_count= b_mb_type_info[mb_type].partition_count;
|
|
3234 mb_type= b_mb_type_info[mb_type].type;
|
|
3235 }else{
|
|
3236 mb_type -= 23;
|
|
3237 goto decode_intra_mb;
|
|
3238 }
|
|
3239 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
|
|
3240 if(mb_type < 5){
|
|
3241 partition_count= p_mb_type_info[mb_type].partition_count;
|
|
3242 mb_type= p_mb_type_info[mb_type].type;
|
|
3243 }else{
|
|
3244 mb_type -= 5;
|
|
3245 goto decode_intra_mb;
|
|
3246 }
|
|
3247 }else{
|
|
3248 assert(h->slice_type == I_TYPE);
|
|
3249 decode_intra_mb:
|
|
3250 if(mb_type > 25){
|
|
3251 fprintf(stderr, "mb_type %d in %c slice to large at %d %d\n", mb_type, ff_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
|
|
3252 return -1;
|
|
3253 }
|
|
3254 partition_count=0;
|
|
3255 cbp= i_mb_type_info[mb_type].cbp;
|
|
3256 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
|
|
3257 mb_type= i_mb_type_info[mb_type].type;
|
|
3258 }
|
|
3259
|
|
3260 if(h->mb_field_decoding_flag)
|
|
3261 mb_type |= MB_TYPE_INTERLACED;
|
|
3262
|
|
3263 s->current_picture.mb_type[mb_xy]= mb_type;
|
|
3264 h->slice_table[ mb_xy ]= h->slice_num;
|
|
3265
|
|
3266 if(IS_INTRA_PCM(mb_type)){
|
|
3267 const uint8_t *ptr;
|
|
3268 int x, y, i;
|
|
3269
|
|
3270 // we assume these blocks are very rare so we dont optimize it
|
|
3271 align_get_bits(&s->gb);
|
|
3272
|
|
3273 ptr= s->gb.buffer + get_bits_count(&s->gb);
|
|
3274
|
|
3275 for(y=0; y<16; y++){
|
|
3276 const int index= 4*(y&3) + 64*(y>>2);
|
|
3277 for(x=0; x<16; x++){
|
|
3278 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
|
|
3279 }
|
|
3280 }
|
|
3281 for(y=0; y<8; y++){
|
|
3282 const int index= 256 + 4*(y&3) + 32*(y>>2);
|
|
3283 for(x=0; x<8; x++){
|
|
3284 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
|
|
3285 }
|
|
3286 }
|
|
3287 for(y=0; y<8; y++){
|
|
3288 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
|
|
3289 for(x=0; x<8; x++){
|
|
3290 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
|
|
3291 }
|
|
3292 }
|
|
3293
|
|
3294 skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
|
|
3295
|
|
3296 memset(h->non_zero_count[mb_xy], 16, 16);
|
|
3297
|
|
3298 return 0;
|
|
3299 }
|
|
3300
|
|
3301 fill_caches(h, mb_type);
|
|
3302
|
|
3303 //mb_pred
|
|
3304 if(IS_INTRA(mb_type)){
|
|
3305 // init_top_left_availability(h);
|
|
3306 if(IS_INTRA4x4(mb_type)){
|
|
3307 int i;
|
|
3308
|
|
3309 // fill_intra4x4_pred_table(h);
|
|
3310 for(i=0; i<16; i++){
|
|
3311 const int mode_coded= !get_bits1(&s->gb);
|
|
3312 const int predicted_mode= pred_intra_mode(h, i);
|
|
3313 int mode;
|
|
3314
|
|
3315 if(mode_coded){
|
|
3316 const int rem_mode= get_bits(&s->gb, 3);
|
|
3317 if(rem_mode<predicted_mode)
|
|
3318 mode= rem_mode;
|
|
3319 else
|
|
3320 mode= rem_mode + 1;
|
|
3321 }else{
|
|
3322 mode= predicted_mode;
|
|
3323 }
|
|
3324
|
|
3325 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
|
|
3326 }
|
|
3327 write_back_intra_pred_mode(h);
|
|
3328 if( check_intra4x4_pred_mode(h) < 0)
|
|
3329 return -1;
|
|
3330 }else{
|
|
3331 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
|
|
3332 if(h->intra16x16_pred_mode < 0)
|
|
3333 return -1;
|
|
3334 }
|
|
3335 h->chroma_pred_mode= get_ue_golomb(&s->gb);
|
|
3336
|
|
3337 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
|
|
3338 if(h->chroma_pred_mode < 0)
|
|
3339 return -1;
|
|
3340 }else if(partition_count==4){
|
|
3341 int i, j, sub_partition_count[4], list, ref[2][4];
|
|
3342
|
|
3343 if(h->slice_type == B_TYPE){
|
|
3344 for(i=0; i<4; i++){
|
|
3345 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
|
|
3346 if(h->sub_mb_type[i] >=13){
|
|
3347 fprintf(stderr, "B sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
|
|
3348 return -1;
|
|
3349 }
|
|
3350 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
|
|
3351 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
|
|
3352 }
|
|
3353 }else{
|
|
3354 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
|
|
3355 for(i=0; i<4; i++){
|
|
3356 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
|
|
3357 if(h->sub_mb_type[i] >=4){
|
|
3358 fprintf(stderr, "P sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
|
|
3359 return -1;
|
|
3360 }
|
|
3361 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
|
|
3362 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
|
|
3363 }
|
|
3364 }
|
|
3365
|
|
3366 for(list=0; list<2; list++){
|
|
3367 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
|
|
3368 if(ref_count == 0) continue;
|
|
3369 for(i=0; i<4; i++){
|
|
3370 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
|
|
3371 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
|
|
3372 }else{
|
|
3373 //FIXME
|
|
3374 ref[list][i] = -1;
|
|
3375 }
|
|
3376 }
|
|
3377 }
|
|
3378
|
|
3379 for(list=0; list<2; list++){
|
|
3380 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
|
|
3381 if(ref_count == 0) continue;
|
|
3382
|
|
3383 for(i=0; i<4; i++){
|
|
3384 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
|
|
3385 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
|
|
3386
|
|
3387 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
|
|
3388 const int sub_mb_type= h->sub_mb_type[i];
|
|
3389 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
|
|
3390 for(j=0; j<sub_partition_count[i]; j++){
|
|
3391 int mx, my;
|
|
3392 const int index= 4*i + block_width*j;
|
|
3393 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
|
|
3394 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
|
|
3395 mx += get_se_golomb(&s->gb);
|
|
3396 my += get_se_golomb(&s->gb);
|
1170
|
3397 tprintf("final mv:%d %d\n", mx, my);
|
|
3398
|
1168
|
3399 if(IS_SUB_8X8(sub_mb_type)){
|
|
3400 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
|
|
3401 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
|
|
3402 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
|
|
3403 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
|
|
3404 }else if(IS_SUB_8X4(sub_mb_type)){
|
|
3405 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
|
|
3406 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
|
|
3407 }else if(IS_SUB_4X8(sub_mb_type)){
|
|
3408 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
|
|
3409 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
|
|
3410 }else{
|
|
3411 assert(IS_SUB_4X4(sub_mb_type));
|
|
3412 mv_cache[ 0 ][0]= mx;
|
|
3413 mv_cache[ 0 ][1]= my;
|
|
3414 }
|
|
3415 }
|
|
3416 }else{
|
|
3417 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
|
|
3418 p[0] = p[1]=
|
|
3419 p[8] = p[9]= 0;
|
|
3420 }
|
|
3421 }
|
|
3422 }
|
|
3423 }else if(!IS_DIRECT(mb_type)){
|
|
3424 int list, mx, my, i;
|
|
3425 //FIXME we should set ref_idx_l? to 0 if we use that later ...
|
|
3426 if(IS_16X16(mb_type)){
|
|
3427 for(list=0; list<2; list++){
|
|
3428 if(h->ref_count[0]>0){
|
|
3429 if(IS_DIR(mb_type, 0, list)){
|
|
3430 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
|
|
3431 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
|
|
3432 }
|
|
3433 }
|
|
3434 }
|
|
3435 for(list=0; list<2; list++){
|
|
3436 if(IS_DIR(mb_type, 0, list)){
|
|
3437 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
|
|
3438 mx += get_se_golomb(&s->gb);
|
|
3439 my += get_se_golomb(&s->gb);
|
1170
|
3440 tprintf("final mv:%d %d\n", mx, my);
|
|
3441
|
1168
|
3442 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, (mx&0xFFFF) + (my<<16), 4);
|
|
3443 }
|
|
3444 }
|
|
3445 }
|
|
3446 else if(IS_16X8(mb_type)){
|
|
3447 for(list=0; list<2; list++){
|
|
3448 if(h->ref_count[list]>0){
|
|
3449 for(i=0; i<2; i++){
|
|
3450 if(IS_DIR(mb_type, i, list)){
|
|
3451 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
|
|
3452 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
|
|
3453 }
|
|
3454 }
|
|
3455 }
|
|
3456 }
|
|
3457 for(list=0; list<2; list++){
|
|
3458 for(i=0; i<2; i++){
|
|
3459 if(IS_DIR(mb_type, i, list)){
|
|
3460 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
|
|
3461 mx += get_se_golomb(&s->gb);
|
|
3462 my += get_se_golomb(&s->gb);
|
1170
|
3463 tprintf("final mv:%d %d\n", mx, my);
|
|
3464
|
1168
|
3465 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (mx&0xFFFF) + (my<<16), 4);
|
|
3466 }
|
|
3467 }
|
|
3468 }
|
|
3469 }else{
|
|
3470 assert(IS_8X16(mb_type));
|
|
3471 for(list=0; list<2; list++){
|
|
3472 if(h->ref_count[list]>0){
|
|
3473 for(i=0; i<2; i++){
|
|
3474 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
|
|
3475 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
|
|
3476 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
|
|
3477 }
|
|
3478 }
|
|
3479 }
|
|
3480 }
|
|
3481 for(list=0; list<2; list++){
|
|
3482 for(i=0; i<2; i++){
|
|
3483 if(IS_DIR(mb_type, i, list)){
|
|
3484 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
|
|
3485 mx += get_se_golomb(&s->gb);
|
|
3486 my += get_se_golomb(&s->gb);
|
1170
|
3487 tprintf("final mv:%d %d\n", mx, my);
|
|
3488
|
1168
|
3489 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (mx&0xFFFF) + (my<<16), 4);
|
|
3490 }
|
|
3491 }
|
|
3492 }
|
|
3493 }
|
|
3494 }
|
|
3495
|
|
3496 if(IS_INTER(mb_type))
|
|
3497 write_back_motion(h, mb_type);
|
|
3498
|
|
3499 if(!IS_INTRA16x16(mb_type)){
|
|
3500 cbp= get_ue_golomb(&s->gb);
|
|
3501 if(cbp > 47){
|
|
3502 fprintf(stderr, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
|
|
3503 return -1;
|
|
3504 }
|
|
3505
|
|
3506 if(IS_INTRA4x4(mb_type))
|
|
3507 cbp= golomb_to_intra4x4_cbp[cbp];
|
|
3508 else
|
|
3509 cbp= golomb_to_inter_cbp[cbp];
|
|
3510 }
|
|
3511
|
|
3512 if(cbp || IS_INTRA16x16(mb_type)){
|
|
3513 int i8x8, i4x4, chroma_idx;
|
|
3514 int chroma_qp, dquant;
|
|
3515 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
|
|
3516 const uint8_t *scan, *dc_scan;
|
|
3517
|
|
3518 // fill_non_zero_count_cache(h);
|
|
3519
|
|
3520 if(IS_INTERLACED(mb_type)){
|
|
3521 scan= field_scan;
|
|
3522 dc_scan= luma_dc_field_scan;
|
|
3523 }else{
|
|
3524 scan= zigzag_scan;
|
|
3525 dc_scan= luma_dc_zigzag_scan;
|
|
3526 }
|
|
3527
|
|
3528 dquant= get_se_golomb(&s->gb);
|
|
3529
|
|
3530 if( dquant > 25 || dquant < -26 ){
|
|
3531 fprintf(stderr, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
|
|
3532 return -1;
|
|
3533 }
|
|
3534
|
|
3535 s->qscale += dquant;
|
|
3536 if(((unsigned)s->qscale) > 51){
|
|
3537 if(s->qscale<0) s->qscale+= 52;
|
|
3538 else s->qscale-= 52;
|
|
3539 }
|
|
3540
|
|
3541 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
|
|
3542 if(IS_INTRA16x16(mb_type)){
|
|
3543 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
|
|
3544 return -1; //FIXME continue if partotioned and other retirn -1 too
|
|
3545 }
|
|
3546
|
|
3547 assert((cbp&15) == 0 || (cbp&15) == 15);
|
|
3548
|
|
3549 if(cbp&15){
|
|
3550 for(i8x8=0; i8x8<4; i8x8++){
|
|
3551 for(i4x4=0; i4x4<4; i4x4++){
|
|
3552 const int index= i4x4 + 4*i8x8;
|
|
3553 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
|
|
3554 return -1;
|
|
3555 }
|
|
3556 }
|
|
3557 }
|
|
3558 }else{
|
|
3559 memset(&h->non_zero_count_cache[8], 0, 8*4); //FIXME stupid & slow
|
|
3560 }
|
|
3561 }else{
|
|
3562 for(i8x8=0; i8x8<4; i8x8++){
|
|
3563 if(cbp & (1<<i8x8)){
|
|
3564 for(i4x4=0; i4x4<4; i4x4++){
|
|
3565 const int index= i4x4 + 4*i8x8;
|
|
3566
|
|
3567 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
|
|
3568 return -1;
|
|
3569 }
|
|
3570 }
|
|
3571 }else{
|
|
3572 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
|
|
3573 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
|
|
3574 }
|
|
3575 }
|
|
3576 }
|
|
3577
|
|
3578 if(cbp&0x30){
|
|
3579 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
|
|
3580 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
|
|
3581 return -1;
|
|
3582 }
|
|
3583 }
|
|
3584
|
|
3585 if(cbp&0x20){
|
|
3586 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
|
|
3587 for(i4x4=0; i4x4<4; i4x4++){
|
|
3588 const int index= 16 + 4*chroma_idx + i4x4;
|
|
3589 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
|
|
3590 return -1;
|
|
3591 }
|
|
3592 }
|
|
3593 }
|
|
3594 }else{
|
|
3595 uint8_t * const nnz= &h->non_zero_count_cache[0];
|
|
3596 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
|
|
3597 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
|
|
3598 }
|
|
3599 }else{
|
|
3600 memset(&h->non_zero_count_cache[8], 0, 8*5);
|
|
3601 }
|
|
3602 write_back_non_zero_count(h);
|
|
3603
|
|
3604 return 0;
|
|
3605 }
|
|
3606
|
|
3607 static int decode_slice(H264Context *h){
|
|
3608 MpegEncContext * const s = &h->s;
|
|
3609 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
|
|
3610
|
|
3611 s->mb_skip_run= -1;
|
|
3612
|
|
3613 #if 1
|
|
3614 for(;;){
|
|
3615 int ret= decode_mb(h);
|
|
3616
|
|
3617 hl_decode_mb(h);
|
|
3618
|
|
3619 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
|
|
3620 s->mb_y++;
|
|
3621 ret= decode_mb(h);
|
|
3622
|
|
3623 hl_decode_mb(h);
|
|
3624 s->mb_y--;
|
|
3625 }
|
|
3626
|
|
3627 if(ret<0){
|
|
3628 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
|
|
3629 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);
|
|
3630
|
|
3631 return -1;
|
|
3632 }
|
|
3633
|
|
3634 if(++s->mb_x >= s->mb_width){
|
|
3635 s->mb_x=0;
|
|
3636 ff_draw_horiz_band(s, 16*s->mb_y, 16);
|
|
3637 if(++s->mb_y >= s->mb_height){
|
1170
|
3638 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
|
|
3639
|
1168
|
3640 if(get_bits_count(&s->gb) == s->gb.size_in_bits){
|
|
3641 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);
|
|
3642
|
|
3643 return 0;
|
|
3644 }else{
|
|
3645 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);
|
|
3646
|
|
3647 return -1;
|
|
3648 }
|
|
3649 }
|
|
3650 }
|
|
3651
|
|
3652 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
|
|
3653 if(get_bits_count(&s->gb) == s->gb.size_in_bits){
|
|
3654 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);
|
|
3655
|
|
3656 return 0;
|
|
3657 }else{
|
|
3658 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);
|
|
3659
|
|
3660 return -1;
|
|
3661 }
|
|
3662 }
|
|
3663 }
|
|
3664 #endif
|
|
3665 #if 0
|
|
3666 for(;s->mb_y < s->mb_height; s->mb_y++){
|
|
3667 for(;s->mb_x < s->mb_width; s->mb_x++){
|
|
3668 int ret= decode_mb(h);
|
|
3669
|
|
3670 hl_decode_mb(h);
|
|
3671
|
|
3672 if(ret<0){
|
|
3673 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
|
|
3674 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);
|
|
3675
|
|
3676 return -1;
|
|
3677 }
|
|
3678
|
|
3679 if(++s->mb_x >= s->mb_width){
|
|
3680 s->mb_x=0;
|
|
3681 if(++s->mb_y >= s->mb_height){
|
|
3682 if(get_bits_count(s->gb) == s->gb.size_in_bits){
|
|
3683 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);
|
|
3684
|
|
3685 return 0;
|
|
3686 }else{
|
|
3687 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);
|
|
3688
|
|
3689 return -1;
|
|
3690 }
|
|
3691 }
|
|
3692 }
|
|
3693
|
|
3694 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
|
|
3695 if(get_bits_count(s->gb) == s->gb.size_in_bits){
|
|
3696 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);
|
|
3697
|
|
3698 return 0;
|
|
3699 }else{
|
|
3700 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);
|
|
3701
|
|
3702 return -1;
|
|
3703 }
|
|
3704 }
|
|
3705 }
|
|
3706 s->mb_x=0;
|
|
3707 ff_draw_horiz_band(s, 16*s->mb_y, 16);
|
|
3708 }
|
|
3709 #endif
|
|
3710 return -1; //not reached
|
|
3711 }
|
|
3712
|
|
3713 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
|
|
3714 MpegEncContext * const s = &h->s;
|
|
3715 int aspect_ratio_info_present_flag, aspect_ratio_idc;
|
|
3716
|
|
3717 aspect_ratio_info_present_flag= get_bits1(&s->gb);
|
|
3718
|
|
3719 if( aspect_ratio_info_present_flag ) {
|
|
3720 aspect_ratio_idc= get_bits(&s->gb, 8);
|
|
3721 if( aspect_ratio_idc == EXTENDED_SAR ) {
|
|
3722 sps->sar_width= get_bits(&s->gb, 16);
|
|
3723 sps->sar_height= get_bits(&s->gb, 16);
|
|
3724 }else if(aspect_ratio_idc < 16){
|
|
3725 sps->sar_width= pixel_aspect[aspect_ratio_idc][0];
|
|
3726 sps->sar_height= pixel_aspect[aspect_ratio_idc][1];
|
|
3727 }else{
|
|
3728 fprintf(stderr, "illegal aspect ratio\n");
|
|
3729 return -1;
|
|
3730 }
|
|
3731 }else{
|
|
3732 sps->sar_width=
|
|
3733 sps->sar_height= 0;
|
|
3734 }
|
|
3735 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
|
|
3736 #if 0
|
|
3737 | overscan_info_present_flag |0 |u(1) |
|
|
3738 | if( overscan_info_present_flag ) | | |
|
|
3739 | overscan_appropriate_flag |0 |u(1) |
|
|
3740 | video_signal_type_present_flag |0 |u(1) |
|
|
3741 | if( video_signal_type_present_flag ) { | | |
|
|
3742 | video_format |0 |u(3) |
|
|
3743 | video_full_range_flag |0 |u(1) |
|
|
3744 | colour_description_present_flag |0 |u(1) |
|
|
3745 | if( colour_description_present_flag ) { | | |
|
|
3746 | colour_primaries |0 |u(8) |
|
|
3747 | transfer_characteristics |0 |u(8) |
|
|
3748 | matrix_coefficients |0 |u(8) |
|
|
3749 | } | | |
|
|
3750 | } | | |
|
|
3751 | chroma_location_info_present_flag |0 |u(1) |
|
|
3752 | if ( chroma_location_info_present_flag ) { | | |
|
|
3753 | chroma_sample_location_type_top_field |0 |ue(v) |
|
|
3754 | chroma_sample_location_type_bottom_field |0 |ue(v) |
|
|
3755 | } | | |
|
|
3756 | timing_info_present_flag |0 |u(1) |
|
|
3757 | if( timing_info_present_flag ) { | | |
|
|
3758 | num_units_in_tick |0 |u(32) |
|
|
3759 | time_scale |0 |u(32) |
|
|
3760 | fixed_frame_rate_flag |0 |u(1) |
|
|
3761 | } | | |
|
|
3762 | nal_hrd_parameters_present_flag |0 |u(1) |
|
|
3763 | if( nal_hrd_parameters_present_flag = = 1) | | |
|
|
3764 | hrd_parameters( ) | | |
|
|
3765 | vcl_hrd_parameters_present_flag |0 |u(1) |
|
|
3766 | if( vcl_hrd_parameters_present_flag = = 1) | | |
|
|
3767 | hrd_parameters( ) | | |
|
|
3768 | if( ( nal_hrd_parameters_present_flag = = 1 | || | |
|
|
3769 | | | |
|
|
3770 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | |
|
|
3771 | low_delay_hrd_flag |0 |u(1) |
|
|
3772 | bitstream_restriction_flag |0 |u(1) |
|
|
3773 | if( bitstream_restriction_flag ) { |0 |u(1) |
|
|
3774 | motion_vectors_over_pic_boundaries_flag |0 |u(1) |
|
|
3775 | max_bytes_per_pic_denom |0 |ue(v) |
|
|
3776 | max_bits_per_mb_denom |0 |ue(v) |
|
|
3777 | log2_max_mv_length_horizontal |0 |ue(v) |
|
|
3778 | log2_max_mv_length_vertical |0 |ue(v) |
|
|
3779 | num_reorder_frames |0 |ue(v) |
|
|
3780 | max_dec_frame_buffering |0 |ue(v) |
|
|
3781 | } | | |
|
|
3782 |} | | |
|
|
3783 #endif
|
|
3784 return 0;
|
|
3785 }
|
|
3786
|
|
3787 static inline int decode_seq_parameter_set(H264Context *h){
|
|
3788 MpegEncContext * const s = &h->s;
|
|
3789 int profile_idc, level_idc, multiple_slice_groups, arbitrary_slice_order, redundant_slices;
|
|
3790 int sps_id, i;
|
|
3791 SPS *sps;
|
|
3792
|
|
3793 profile_idc= get_bits(&s->gb, 8);
|
|
3794 level_idc= get_bits(&s->gb, 8);
|
|
3795 multiple_slice_groups= get_bits1(&s->gb);
|
|
3796 arbitrary_slice_order= get_bits1(&s->gb);
|
|
3797 redundant_slices= get_bits1(&s->gb);
|
|
3798
|
|
3799 sps_id= get_ue_golomb(&s->gb);
|
|
3800
|
|
3801 sps= &h->sps_buffer[ sps_id ];
|
|
3802
|
|
3803 sps->profile_idc= profile_idc;
|
|
3804 sps->level_idc= level_idc;
|
|
3805 sps->multiple_slice_groups= multiple_slice_groups;
|
|
3806 sps->arbitrary_slice_order= arbitrary_slice_order;
|
|
3807 sps->redundant_slices= redundant_slices;
|
|
3808
|
|
3809 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
|
|
3810
|
|
3811 sps->poc_type= get_ue_golomb(&s->gb);
|
|
3812
|
|
3813 if(sps->poc_type == 0){ //FIXME #define
|
|
3814 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
|
|
3815 } else if(sps->poc_type == 1){//FIXME #define
|
|
3816 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
|
|
3817 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
|
|
3818 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
|
|
3819 sps->poc_cycle_length= get_ue_golomb(&s->gb);
|
|
3820
|
|
3821 for(i=0; i<sps->poc_cycle_length; i++)
|
|
3822 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
|
|
3823 }
|
|
3824 if(sps->poc_type > 2){
|
|
3825 fprintf(stderr, "illegal POC type %d\n", sps->poc_type);
|
|
3826 return -1;
|
|
3827 }
|
|
3828
|
|
3829 sps->ref_frame_count= get_ue_golomb(&s->gb);
|
|
3830 sps->required_frame_num_update_behaviour_flag= get_bits1(&s->gb);
|
|
3831 sps->mb_width= get_ue_golomb(&s->gb) + 1;
|
|
3832 sps->mb_height= get_ue_golomb(&s->gb) + 1;
|
|
3833 sps->frame_mbs_only_flag= get_bits1(&s->gb);
|
|
3834 if(!sps->frame_mbs_only_flag)
|
|
3835 sps->mb_aff= get_bits1(&s->gb);
|
|
3836 else
|
|
3837 sps->mb_aff= 0;
|
|
3838
|
|
3839 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
|
|
3840
|
|
3841 sps->vui_parameters_present_flag= get_bits1(&s->gb);
|
|
3842 if( sps->vui_parameters_present_flag )
|
|
3843 decode_vui_parameters(h, sps);
|
|
3844
|
|
3845 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
|
|
3846 printf("sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s %s\n",
|
|
3847 sps_id, sps->profile_idc, sps->level_idc,
|
|
3848 sps->poc_type,
|
|
3849 sps->ref_frame_count,
|
|
3850 sps->mb_width, sps->mb_height,
|
|
3851 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
|
|
3852 sps->direct_8x8_inference_flag ? "8B8" : "",
|
|
3853 sps->vui_parameters_present_flag ? "VUI" : ""
|
|
3854 );
|
|
3855 }
|
|
3856 return 0;
|
|
3857 }
|
|
3858
|
|
3859 static inline int decode_picture_parameter_set(H264Context *h){
|
|
3860 MpegEncContext * const s = &h->s;
|
|
3861 int pps_id= get_ue_golomb(&s->gb);
|
|
3862 PPS *pps= &h->pps_buffer[pps_id];
|
|
3863
|
|
3864 pps->sps_id= get_ue_golomb(&s->gb);
|
|
3865 pps->cabac= get_bits1(&s->gb);
|
|
3866 pps->pic_order_present= get_bits1(&s->gb);
|
|
3867 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
|
|
3868 if(pps->slice_group_count > 1 ){
|
|
3869 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
|
|
3870 fprintf(stderr, "FMO not supported\n");
|
|
3871 switch(pps->mb_slice_group_map_type){
|
|
3872 case 0:
|
|
3873 #if 0
|
|
3874 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
|
|
3875 | run_length[ i ] |1 |ue(v) |
|
|
3876 #endif
|
|
3877 break;
|
|
3878 case 2:
|
|
3879 #if 0
|
|
3880 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
|
|
3881 |{ | | |
|
|
3882 | top_left_mb[ i ] |1 |ue(v) |
|
|
3883 | bottom_right_mb[ i ] |1 |ue(v) |
|
|
3884 | } | | |
|
|
3885 #endif
|
|
3886 break;
|
|
3887 case 3:
|
|
3888 case 4:
|
|
3889 case 5:
|
|
3890 #if 0
|
|
3891 | slice_group_change_direction_flag |1 |u(1) |
|
|
3892 | slice_group_change_rate_minus1 |1 |ue(v) |
|
|
3893 #endif
|
|
3894 break;
|
|
3895 case 6:
|
|
3896 #if 0
|
|
3897 | slice_group_id_cnt_minus1 |1 |ue(v) |
|
|
3898 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
|
|
3899 |) | | |
|
|
3900 | slice_group_id[ i ] |1 |u(v) |
|
|
3901 #endif
|
|
3902 }
|
|
3903 }
|
|
3904 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
|
|
3905 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
|
|
3906 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
|
|
3907 fprintf(stderr, "reference overflow (pps)\n");
|
|
3908 return -1;
|
|
3909 }
|
|
3910
|
|
3911 pps->weighted_pred= get_bits1(&s->gb);
|
|
3912 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
|
|
3913 pps->init_qp= get_se_golomb(&s->gb) + 26;
|
|
3914 pps->init_qs= get_se_golomb(&s->gb) + 26;
|
|
3915 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
|
|
3916 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
|
|
3917 pps->constrained_intra_pred= get_bits1(&s->gb);
|
|
3918 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
|
|
3919 pps->crop= get_bits1(&s->gb);
|
|
3920 if(pps->crop){
|
|
3921 pps->crop_left = get_ue_golomb(&s->gb);
|
|
3922 pps->crop_right = get_ue_golomb(&s->gb);
|
|
3923 pps->crop_top = get_ue_golomb(&s->gb);
|
|
3924 pps->crop_bottom= get_ue_golomb(&s->gb);
|
|
3925 }else{
|
|
3926 pps->crop_left =
|
|
3927 pps->crop_right =
|
|
3928 pps->crop_top =
|
|
3929 pps->crop_bottom= 0;
|
|
3930 }
|
|
3931
|
|
3932 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
|
|
3933 printf("pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s crop:%d/%d/%d/%d\n",
|
|
3934 pps_id, pps->sps_id,
|
|
3935 pps->cabac ? "CABAC" : "CAVLC",
|
|
3936 pps->slice_group_count,
|
|
3937 pps->ref_count[0], pps->ref_count[1],
|
|
3938 pps->weighted_pred ? "weighted" : "",
|
|
3939 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
|
|
3940 pps->deblocking_filter_parameters_present ? "LPAR" : "",
|
|
3941 pps->constrained_intra_pred ? "CONSTR" : "",
|
|
3942 pps->redundant_pic_cnt_present ? "REDU" : "",
|
|
3943 pps->crop_left, pps->crop_right,
|
|
3944 pps->crop_top, pps->crop_bottom
|
|
3945 );
|
|
3946 }
|
|
3947
|
|
3948 return 0;
|
|
3949 }
|
|
3950
|
|
3951 /**
|
|
3952 * finds the end of the current frame in the bitstream.
|
|
3953 * @return the position of the first byte of the next frame, or -1
|
|
3954 */
|
|
3955 static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){
|
|
3956 ParseContext *pc= &s->parse_context;
|
|
3957 int last_addr, i;
|
|
3958 uint32_t state;
|
|
3959 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
|
|
3960 // mb_addr= pc->mb_addr - 1;
|
|
3961 state= pc->state;
|
|
3962 //FIXME this will fail with slices
|
|
3963 for(i=0; i<buf_size; i++){
|
|
3964 state= (state<<8) | buf[i];
|
|
3965 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
|
|
3966 if(pc->frame_start_found){
|
|
3967 pc->state=-1;
|
|
3968 pc->frame_start_found= 0;
|
|
3969 return i-3;
|
|
3970 }
|
|
3971 pc->frame_start_found= 1;
|
|
3972 }
|
|
3973 }
|
|
3974
|
|
3975 pc->state= state;
|
|
3976 return -1;
|
|
3977 }
|
|
3978
|
|
3979 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
|
|
3980 MpegEncContext * const s = &h->s;
|
|
3981 AVCodecContext * const avctx= s->avctx;
|
|
3982 int buf_index=0;
|
|
3983 int i;
|
|
3984 #if 0
|
|
3985 for(i=0; i<32; i++){
|
|
3986 printf("%X ", buf[i]);
|
|
3987 }
|
|
3988 #endif
|
|
3989 for(;;){
|
|
3990 int consumed;
|
|
3991 int dst_length;
|
|
3992 int bit_length;
|
|
3993 uint8_t *ptr;
|
|
3994
|
|
3995 // start code prefix search
|
|
3996 for(; buf_index + 3 < buf_size; buf_index++){
|
|
3997 // this should allways succeed in the first iteration
|
|
3998 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
|
|
3999 break;
|
|
4000 }
|
|
4001
|
|
4002 if(buf_index+3 >= buf_size) break;
|
|
4003
|
|
4004 buf_index+=3;
|
|
4005
|
|
4006 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
|
|
4007 if(ptr[dst_length - 1] == 0) dst_length--;
|
|
4008 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
|
|
4009
|
|
4010 if(s->avctx->debug&FF_DEBUG_STARTCODE){
|
|
4011 printf("NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
|
|
4012 }
|
|
4013
|
|
4014 buf_index += consumed;
|
|
4015
|
|
4016 if(h->nal_ref_idc < s->hurry_up)
|
|
4017 continue;
|
|
4018
|
|
4019 switch(h->nal_unit_type){
|
|
4020 case NAL_IDR_SLICE:
|
|
4021 idr(h); //FIXME ensure we dont loose some frames if there is reordering
|
|
4022 case NAL_SLICE:
|
|
4023 init_get_bits(&s->gb, ptr, bit_length);
|
|
4024 h->intra_gb_ptr=
|
|
4025 h->inter_gb_ptr= &s->gb;
|
|
4026 s->data_partitioning = 0;
|
|
4027
|
|
4028 if(decode_slice_header(h) < 0) return -1;
|
|
4029 if(h->redundant_pic_count==0)
|
|
4030 decode_slice(h);
|
|
4031 break;
|
|
4032 case NAL_DPA:
|
|
4033 init_get_bits(&s->gb, ptr, bit_length);
|
|
4034 h->intra_gb_ptr=
|
|
4035 h->inter_gb_ptr= NULL;
|
|
4036 s->data_partitioning = 1;
|
|
4037
|
|
4038 if(decode_slice_header(h) < 0) return -1;
|
|
4039 break;
|
|
4040 case NAL_DPB:
|
|
4041 init_get_bits(&h->intra_gb, ptr, bit_length);
|
|
4042 h->intra_gb_ptr= &h->intra_gb;
|
|
4043 break;
|
|
4044 case NAL_DPC:
|
|
4045 init_get_bits(&h->inter_gb, ptr, bit_length);
|
|
4046 h->inter_gb_ptr= &h->inter_gb;
|
1174
|
4047
|
|
4048 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning)
|
1168
|
4049 decode_slice(h);
|
|
4050 break;
|
|
4051 case NAL_SEI:
|
|
4052 break;
|
|
4053 case NAL_SPS:
|
|
4054 init_get_bits(&s->gb, ptr, bit_length);
|
|
4055 decode_seq_parameter_set(h);
|
|
4056
|
|
4057 if(s->flags& CODEC_FLAG_LOW_DELAY)
|
|
4058 s->low_delay=1;
|
|
4059
|
|
4060 avctx->has_b_frames= !s->low_delay;
|
|
4061 break;
|
|
4062 case NAL_PPS:
|
|
4063 init_get_bits(&s->gb, ptr, bit_length);
|
|
4064
|
|
4065 decode_picture_parameter_set(h);
|
|
4066
|
|
4067 break;
|
|
4068 case NAL_PICTURE_DELIMITER:
|
|
4069 break;
|
|
4070 case NAL_FILTER_DATA:
|
|
4071 break;
|
|
4072 }
|
|
4073
|
|
4074 //FIXME move after where irt is set
|
|
4075 s->current_picture.pict_type= s->pict_type;
|
|
4076 s->current_picture.key_frame= s->pict_type == I_TYPE;
|
|
4077 }
|
|
4078
|
1174
|
4079 if(!s->current_picture_ptr) return buf_index; //no frame
|
|
4080
|
1168
|
4081 h->prev_frame_num_offset= h->frame_num_offset;
|
|
4082 h->prev_frame_num= h->frame_num;
|
|
4083 if(s->current_picture_ptr->reference){
|
|
4084 h->prev_poc_msb= h->poc_msb;
|
|
4085 h->prev_poc_lsb= h->poc_lsb;
|
|
4086 }
|
|
4087 if(s->current_picture_ptr->reference)
|
|
4088 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
|
|
4089 else
|
1169
|
4090 assert(h->mmco_index==0);
|
1168
|
4091
|
|
4092 ff_er_frame_end(s);
|
|
4093 MPV_frame_end(s);
|
|
4094
|
|
4095 return buf_index;
|
|
4096 }
|
|
4097
|
|
4098 /**
|
|
4099 * retunrs the number of bytes consumed for building the current frame
|
|
4100 */
|
|
4101 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
|
|
4102 if(s->flags&CODEC_FLAG_TRUNCATED){
|
|
4103 pos -= s->parse_context.last_index;
|
|
4104 if(pos<0) pos=0; // FIXME remove (uneeded?)
|
|
4105
|
|
4106 return pos;
|
|
4107 }else{
|
|
4108 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
|
|
4109 if(pos+10>buf_size) pos=buf_size; // oops ;)
|
|
4110
|
|
4111 return pos;
|
|
4112 }
|
|
4113 }
|
|
4114
|
|
4115 static int decode_frame(AVCodecContext *avctx,
|
|
4116 void *data, int *data_size,
|
|
4117 uint8_t *buf, int buf_size)
|
|
4118 {
|
|
4119 H264Context *h = avctx->priv_data;
|
|
4120 MpegEncContext *s = &h->s;
|
|
4121 AVFrame *pict = data;
|
|
4122 float new_aspect;
|
|
4123 int buf_index;
|
|
4124
|
|
4125 s->flags= avctx->flags;
|
|
4126
|
|
4127 *data_size = 0;
|
|
4128
|
|
4129 /* no supplementary picture */
|
|
4130 if (buf_size == 0) {
|
|
4131 return 0;
|
|
4132 }
|
|
4133
|
|
4134 if(s->flags&CODEC_FLAG_TRUNCATED){
|
|
4135 int next= find_frame_end(s, buf, buf_size);
|
|
4136
|
|
4137 if( ff_combine_frame(s, next, &buf, &buf_size) < 0 )
|
|
4138 return buf_size;
|
|
4139 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
|
|
4140 }
|
|
4141
|
|
4142 if(s->avctx->extradata_size && s->picture_number==0){
|
|
4143 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
|
|
4144 return -1;
|
|
4145 }
|
|
4146
|
|
4147 buf_index=decode_nal_units(h, buf, buf_size);
|
|
4148 if(buf_index < 0)
|
|
4149 return -1;
|
|
4150
|
|
4151 //FIXME do something with unavailable reference frames
|
|
4152
|
|
4153 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
|
|
4154 #if 0
|
|
4155 if(s->pict_type==B_TYPE || s->low_delay){
|
|
4156 *pict= *(AVFrame*)&s->current_picture;
|
|
4157 } else {
|
|
4158 *pict= *(AVFrame*)&s->last_picture;
|
|
4159 }
|
|
4160 #endif
|
1174
|
4161 if(!s->current_picture_ptr){
|
|
4162 fprintf(stderr, "error, NO frame\n");
|
|
4163 return -1;
|
|
4164 }
|
|
4165
|
1168
|
4166 *pict= *(AVFrame*)&s->current_picture; //FIXME
|
|
4167 assert(pict->data[0]);
|
|
4168 //printf("out %d\n", (int)pict->data[0]);
|
|
4169 #if 0 //?
|
|
4170
|
|
4171 /* Return the Picture timestamp as the frame number */
|
|
4172 /* we substract 1 because it is added on utils.c */
|
|
4173 avctx->frame_number = s->picture_number - 1;
|
|
4174 #endif
|
|
4175 #if 0
|
|
4176 /* dont output the last pic after seeking */
|
|
4177 if(s->last_picture_ptr || s->low_delay)
|
|
4178 //Note this isnt a issue as a IDR pic should flush teh buffers
|
|
4179 #endif
|
|
4180 *data_size = sizeof(AVFrame);
|
|
4181 return get_consumed_bytes(s, buf_index, buf_size);
|
|
4182 }
|
|
4183 #if 0
|
|
4184 static inline void fill_mb_avail(H264Context *h){
|
|
4185 MpegEncContext * const s = &h->s;
|
|
4186 const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
|
|
4187
|
|
4188 if(s->mb_y){
|
|
4189 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - h->mb_stride - 1] == h->slice_num;
|
|
4190 h->mb_avail[1]= h->slice_table[mb_xy - h->mb_stride ] == h->slice_num;
|
|
4191 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - h->mb_stride + 1] == h->slice_num;
|
|
4192 }else{
|
|
4193 h->mb_avail[0]=
|
|
4194 h->mb_avail[1]=
|
|
4195 h->mb_avail[2]= 0;
|
|
4196 }
|
|
4197 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
|
|
4198 h->mb_avail[4]= 1; //FIXME move out
|
|
4199 h->mb_avail[5]= 0; //FIXME move out
|
|
4200 }
|
|
4201 #endif
|
|
4202
|
|
4203 #if 0 //selftest
|
|
4204 #define COUNT 8000
|
|
4205 #define SIZE (COUNT*40)
|
|
4206 int main(){
|
|
4207 int i;
|
|
4208 uint8_t temp[SIZE];
|
|
4209 PutBitContext pb;
|
|
4210 GetBitContext gb;
|
|
4211 // int int_temp[10000];
|
|
4212 DSPContext dsp;
|
|
4213 AVCodecContext avctx;
|
|
4214
|
|
4215 dsputil_init(&dsp, &avctx);
|
|
4216
|
|
4217 init_put_bits(&pb, temp, SIZE, NULL, NULL);
|
|
4218 printf("testing unsigned exp golomb\n");
|
|
4219 for(i=0; i<COUNT; i++){
|
|
4220 START_TIMER
|
|
4221 set_ue_golomb(&pb, i);
|
|
4222 STOP_TIMER("set_ue_golomb");
|
|
4223 }
|
|
4224 flush_put_bits(&pb);
|
|
4225
|
|
4226 init_get_bits(&gb, temp, 8*SIZE);
|
|
4227 for(i=0; i<COUNT; i++){
|
|
4228 int j, s;
|
|
4229
|
|
4230 s= show_bits(&gb, 24);
|
|
4231
|
|
4232 START_TIMER
|
|
4233 j= get_ue_golomb(&gb);
|
|
4234 if(j != i){
|
|
4235 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
|
|
4236 // return -1;
|
|
4237 }
|
|
4238 STOP_TIMER("get_ue_golomb");
|
|
4239 }
|
|
4240
|
|
4241
|
|
4242 init_put_bits(&pb, temp, SIZE, NULL, NULL);
|
|
4243 printf("testing signed exp golomb\n");
|
|
4244 for(i=0; i<COUNT; i++){
|
|
4245 START_TIMER
|
|
4246 set_se_golomb(&pb, i - COUNT/2);
|
|
4247 STOP_TIMER("set_se_golomb");
|
|
4248 }
|
|
4249 flush_put_bits(&pb);
|
|
4250
|
|
4251 init_get_bits(&gb, temp, 8*SIZE);
|
|
4252 for(i=0; i<COUNT; i++){
|
|
4253 int j, s;
|
|
4254
|
|
4255 s= show_bits(&gb, 24);
|
|
4256
|
|
4257 START_TIMER
|
|
4258 j= get_se_golomb(&gb);
|
|
4259 if(j != i - COUNT/2){
|
|
4260 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
|
|
4261 // return -1;
|
|
4262 }
|
|
4263 STOP_TIMER("get_se_golomb");
|
|
4264 }
|
|
4265
|
|
4266 printf("testing 4x4 (I)DCT\n");
|
|
4267
|
|
4268 DCTELEM block[16];
|
|
4269 uint8_t src[16], ref[16];
|
|
4270 uint64_t error= 0, max_error=0;
|
|
4271
|
|
4272 for(i=0; i<COUNT; i++){
|
|
4273 int j;
|
|
4274 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
|
|
4275 for(j=0; j<16; j++){
|
|
4276 ref[j]= random()%255;
|
|
4277 src[j]= random()%255;
|
|
4278 }
|
|
4279
|
|
4280 h264_diff_dct_c(block, src, ref, 4);
|
|
4281
|
|
4282 //normalize
|
|
4283 for(j=0; j<16; j++){
|
|
4284 // printf("%d ", block[j]);
|
|
4285 block[j]= block[j]*4;
|
|
4286 if(j&1) block[j]= (block[j]*4 + 2)/5;
|
|
4287 if(j&4) block[j]= (block[j]*4 + 2)/5;
|
|
4288 }
|
|
4289 // printf("\n");
|
|
4290
|
|
4291 h264_add_idct_c(ref, block, 4);
|
|
4292 /* for(j=0; j<16; j++){
|
|
4293 printf("%d ", ref[j]);
|
|
4294 }
|
|
4295 printf("\n");*/
|
|
4296
|
|
4297 for(j=0; j<16; j++){
|
|
4298 int diff= ABS(src[j] - ref[j]);
|
|
4299
|
|
4300 error+= diff*diff;
|
|
4301 max_error= FFMAX(max_error, diff);
|
|
4302 }
|
|
4303 }
|
|
4304 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
|
|
4305 #if 0
|
|
4306 printf("testing quantizer\n");
|
|
4307 for(qp=0; qp<52; qp++){
|
|
4308 for(i=0; i<16; i++)
|
|
4309 src1_block[i]= src2_block[i]= random()%255;
|
|
4310
|
|
4311 }
|
|
4312 #endif
|
|
4313 printf("Testing NAL layer\n");
|
|
4314
|
|
4315 uint8_t bitstream[COUNT];
|
|
4316 uint8_t nal[COUNT*2];
|
|
4317 H264Context h;
|
|
4318 memset(&h, 0, sizeof(H264Context));
|
|
4319
|
|
4320 for(i=0; i<COUNT; i++){
|
|
4321 int zeros= i;
|
|
4322 int nal_length;
|
|
4323 int consumed;
|
|
4324 int out_length;
|
|
4325 uint8_t *out;
|
|
4326 int j;
|
|
4327
|
|
4328 for(j=0; j<COUNT; j++){
|
|
4329 bitstream[j]= (random() % 255) + 1;
|
|
4330 }
|
|
4331
|
|
4332 for(j=0; j<zeros; j++){
|
|
4333 int pos= random() % COUNT;
|
|
4334 while(bitstream[pos] == 0){
|
|
4335 pos++;
|
|
4336 pos %= COUNT;
|
|
4337 }
|
|
4338 bitstream[pos]=0;
|
|
4339 }
|
|
4340
|
|
4341 START_TIMER
|
|
4342
|
|
4343 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
|
|
4344 if(nal_length<0){
|
|
4345 printf("encoding failed\n");
|
|
4346 return -1;
|
|
4347 }
|
|
4348
|
|
4349 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
|
|
4350
|
|
4351 STOP_TIMER("NAL")
|
|
4352
|
|
4353 if(out_length != COUNT){
|
|
4354 printf("incorrect length %d %d\n", out_length, COUNT);
|
|
4355 return -1;
|
|
4356 }
|
|
4357
|
|
4358 if(consumed != nal_length){
|
|
4359 printf("incorrect consumed length %d %d\n", nal_length, consumed);
|
|
4360 return -1;
|
|
4361 }
|
|
4362
|
|
4363 if(memcmp(bitstream, out, COUNT)){
|
|
4364 printf("missmatch\n");
|
|
4365 return -1;
|
|
4366 }
|
|
4367 }
|
|
4368
|
|
4369 printf("Testing RBSP\n");
|
|
4370
|
|
4371
|
|
4372 return 0;
|
|
4373 }
|
|
4374 #endif
|
|
4375
|
|
4376
|
|
4377 static int decode_end(AVCodecContext *avctx)
|
|
4378 {
|
|
4379 H264Context *h = avctx->priv_data;
|
|
4380 MpegEncContext *s = &h->s;
|
|
4381
|
|
4382 free_tables(h); //FIXME cleanup init stuff perhaps
|
|
4383 MPV_common_end(s);
|
|
4384
|
|
4385 // memset(h, 0, sizeof(H264Context));
|
|
4386
|
|
4387 return 0;
|
|
4388 }
|
|
4389
|
|
4390
|
|
4391 AVCodec h264_decoder = {
|
|
4392 "h264",
|
|
4393 CODEC_TYPE_VIDEO,
|
|
4394 CODEC_ID_H264,
|
|
4395 sizeof(H264Context),
|
|
4396 decode_init,
|
|
4397 NULL,
|
|
4398 decode_end,
|
|
4399 decode_frame,
|
|
4400 /*CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_DR1 | */CODEC_CAP_TRUNCATED,
|
|
4401 };
|
|
4402
|