Fix svq3_* function declarations.
[libav.git] / libavcodec / h264.c
1 /*
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file libavcodec/h264.c
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28 #include "internal.h"
29 #include "dsputil.h"
30 #include "avcodec.h"
31 #include "mpegvideo.h"
32 #include "h264.h"
33 #include "h264data.h"
34 #include "h264_mvpred.h"
35 #include "h264_parser.h"
36 #include "golomb.h"
37 #include "mathops.h"
38 #include "rectangle.h"
39 #include "vdpau_internal.h"
40
41 #include "cabac.h"
42
43 //#undef NDEBUG
44 #include <assert.h>
45
46 static const uint8_t rem6[52]={
47 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
48 };
49
50 static const uint8_t div6[52]={
51 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
52 };
53
54 void ff_h264_write_back_intra_pred_mode(H264Context *h){
55 const int mb_xy= h->mb_xy;
56
57 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
58 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
59 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
60 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
61 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
62 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
63 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
64 }
65
66 /**
67 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
68 */
69 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
70 MpegEncContext * const s = &h->s;
71 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
72 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
73 int i;
74
75 if(!(h->top_samples_available&0x8000)){
76 for(i=0; i<4; i++){
77 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
78 if(status<0){
79 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
80 return -1;
81 } else if(status){
82 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
83 }
84 }
85 }
86
87 if((h->left_samples_available&0x8888)!=0x8888){
88 static const int mask[4]={0x8000,0x2000,0x80,0x20};
89 for(i=0; i<4; i++){
90 if(!(h->left_samples_available&mask[i])){
91 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
92 if(status<0){
93 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
94 return -1;
95 } else if(status){
96 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
97 }
98 }
99 }
100 }
101
102 return 0;
103 } //FIXME cleanup like ff_h264_check_intra_pred_mode
104
105 /**
106 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
107 */
108 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
109 MpegEncContext * const s = &h->s;
110 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
111 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
112
113 if(mode > 6U) {
114 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
115 return -1;
116 }
117
118 if(!(h->top_samples_available&0x8000)){
119 mode= top[ mode ];
120 if(mode<0){
121 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
122 return -1;
123 }
124 }
125
126 if((h->left_samples_available&0x8080) != 0x8080){
127 mode= left[ mode ];
128 if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
129 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
130 }
131 if(mode<0){
132 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
133 return -1;
134 }
135 }
136
137 return mode;
138 }
139
140 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
141 int i, si, di;
142 uint8_t *dst;
143 int bufidx;
144
145 // src[0]&0x80; //forbidden bit
146 h->nal_ref_idc= src[0]>>5;
147 h->nal_unit_type= src[0]&0x1F;
148
149 src++; length--;
150 #if 0
151 for(i=0; i<length; i++)
152 printf("%2X ", src[i]);
153 #endif
154
155 #if HAVE_FAST_UNALIGNED
156 # if HAVE_FAST_64BIT
157 # define RS 7
158 for(i=0; i+1<length; i+=9){
159 if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
160 # else
161 # define RS 3
162 for(i=0; i+1<length; i+=5){
163 if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
164 # endif
165 continue;
166 if(i>0 && !src[i]) i--;
167 while(src[i]) i++;
168 #else
169 # define RS 0
170 for(i=0; i+1<length; i+=2){
171 if(src[i]) continue;
172 if(i>0 && src[i-1]==0) i--;
173 #endif
174 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
175 if(src[i+2]!=3){
176 /* startcode, so we must be past the end */
177 length=i;
178 }
179 break;
180 }
181 i-= RS;
182 }
183
184 if(i>=length-1){ //no escaped 0
185 *dst_length= length;
186 *consumed= length+1; //+1 for the header
187 return src;
188 }
189
190 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
191 av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
192 dst= h->rbsp_buffer[bufidx];
193
194 if (dst == NULL){
195 return NULL;
196 }
197
198 //printf("decoding esc\n");
199 memcpy(dst, src, i);
200 si=di=i;
201 while(si+2<length){
202 //remove escapes (very rare 1:2^22)
203 if(src[si+2]>3){
204 dst[di++]= src[si++];
205 dst[di++]= src[si++];
206 }else if(src[si]==0 && src[si+1]==0){
207 if(src[si+2]==3){ //escape
208 dst[di++]= 0;
209 dst[di++]= 0;
210 si+=3;
211 continue;
212 }else //next start code
213 goto nsc;
214 }
215
216 dst[di++]= src[si++];
217 }
218 while(si<length)
219 dst[di++]= src[si++];
220 nsc:
221
222 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
223
224 *dst_length= di;
225 *consumed= si + 1;//+1 for the header
226 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
227 return dst;
228 }
229
230 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
231 int v= *src;
232 int r;
233
234 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
235
236 for(r=1; r<9; r++){
237 if(v&1) return r;
238 v>>=1;
239 }
240 return 0;
241 }
242
243 /**
244 * IDCT transforms the 16 dc values and dequantizes them.
245 * @param qp quantization parameter
246 */
247 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
248 #define stride 16
249 int i;
250 int temp[16]; //FIXME check if this is a good idea
251 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
252 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
253
254 //memset(block, 64, 2*256);
255 //return;
256 for(i=0; i<4; i++){
257 const int offset= y_offset[i];
258 const int z0= block[offset+stride*0] + block[offset+stride*4];
259 const int z1= block[offset+stride*0] - block[offset+stride*4];
260 const int z2= block[offset+stride*1] - block[offset+stride*5];
261 const int z3= block[offset+stride*1] + block[offset+stride*5];
262
263 temp[4*i+0]= z0+z3;
264 temp[4*i+1]= z1+z2;
265 temp[4*i+2]= z1-z2;
266 temp[4*i+3]= z0-z3;
267 }
268
269 for(i=0; i<4; i++){
270 const int offset= x_offset[i];
271 const int z0= temp[4*0+i] + temp[4*2+i];
272 const int z1= temp[4*0+i] - temp[4*2+i];
273 const int z2= temp[4*1+i] - temp[4*3+i];
274 const int z3= temp[4*1+i] + temp[4*3+i];
275
276 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
277 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
278 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
279 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
280 }
281 }
282
283 #if 0
284 /**
285 * DCT transforms the 16 dc values.
286 * @param qp quantization parameter ??? FIXME
287 */
288 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
289 // const int qmul= dequant_coeff[qp][0];
290 int i;
291 int temp[16]; //FIXME check if this is a good idea
292 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
293 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
294
295 for(i=0; i<4; i++){
296 const int offset= y_offset[i];
297 const int z0= block[offset+stride*0] + block[offset+stride*4];
298 const int z1= block[offset+stride*0] - block[offset+stride*4];
299 const int z2= block[offset+stride*1] - block[offset+stride*5];
300 const int z3= block[offset+stride*1] + block[offset+stride*5];
301
302 temp[4*i+0]= z0+z3;
303 temp[4*i+1]= z1+z2;
304 temp[4*i+2]= z1-z2;
305 temp[4*i+3]= z0-z3;
306 }
307
308 for(i=0; i<4; i++){
309 const int offset= x_offset[i];
310 const int z0= temp[4*0+i] + temp[4*2+i];
311 const int z1= temp[4*0+i] - temp[4*2+i];
312 const int z2= temp[4*1+i] - temp[4*3+i];
313 const int z3= temp[4*1+i] + temp[4*3+i];
314
315 block[stride*0 +offset]= (z0 + z3)>>1;
316 block[stride*2 +offset]= (z1 + z2)>>1;
317 block[stride*8 +offset]= (z1 - z2)>>1;
318 block[stride*10+offset]= (z0 - z3)>>1;
319 }
320 }
321 #endif
322
323 #undef xStride
324 #undef stride
325
326 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
327 const int stride= 16*2;
328 const int xStride= 16;
329 int a,b,c,d,e;
330
331 a= block[stride*0 + xStride*0];
332 b= block[stride*0 + xStride*1];
333 c= block[stride*1 + xStride*0];
334 d= block[stride*1 + xStride*1];
335
336 e= a-b;
337 a= a+b;
338 b= c-d;
339 c= c+d;
340
341 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
342 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
343 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
344 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
345 }
346
347 #if 0
348 static void chroma_dc_dct_c(DCTELEM *block){
349 const int stride= 16*2;
350 const int xStride= 16;
351 int a,b,c,d,e;
352
353 a= block[stride*0 + xStride*0];
354 b= block[stride*0 + xStride*1];
355 c= block[stride*1 + xStride*0];
356 d= block[stride*1 + xStride*1];
357
358 e= a-b;
359 a= a+b;
360 b= c-d;
361 c= c+d;
362
363 block[stride*0 + xStride*0]= (a+c);
364 block[stride*0 + xStride*1]= (e+b);
365 block[stride*1 + xStride*0]= (a-c);
366 block[stride*1 + xStride*1]= (e-b);
367 }
368 #endif
369
370 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
371 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
372 int src_x_offset, int src_y_offset,
373 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
374 MpegEncContext * const s = &h->s;
375 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
376 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
377 const int luma_xy= (mx&3) + ((my&3)<<2);
378 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
379 uint8_t * src_cb, * src_cr;
380 int extra_width= h->emu_edge_width;
381 int extra_height= h->emu_edge_height;
382 int emu=0;
383 const int full_mx= mx>>2;
384 const int full_my= my>>2;
385 const int pic_width = 16*s->mb_width;
386 const int pic_height = 16*s->mb_height >> MB_FIELD;
387
388 if(mx&7) extra_width -= 3;
389 if(my&7) extra_height -= 3;
390
391 if( full_mx < 0-extra_width
392 || full_my < 0-extra_height
393 || full_mx + 16/*FIXME*/ > pic_width + extra_width
394 || full_my + 16/*FIXME*/ > pic_height + extra_height){
395 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
396 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
397 emu=1;
398 }
399
400 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
401 if(!square){
402 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
403 }
404
405 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
406
407 if(MB_FIELD){
408 // chroma offset when predicting from a field of opposite parity
409 my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
410 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
411 }
412 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
413 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
414
415 if(emu){
416 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
417 src_cb= s->edge_emu_buffer;
418 }
419 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
420
421 if(emu){
422 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
423 src_cr= s->edge_emu_buffer;
424 }
425 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
426 }
427
428 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
429 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
430 int x_offset, int y_offset,
431 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
432 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
433 int list0, int list1){
434 MpegEncContext * const s = &h->s;
435 qpel_mc_func *qpix_op= qpix_put;
436 h264_chroma_mc_func chroma_op= chroma_put;
437
438 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
439 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
440 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
441 x_offset += 8*s->mb_x;
442 y_offset += 8*(s->mb_y >> MB_FIELD);
443
444 if(list0){
445 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
446 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
447 dest_y, dest_cb, dest_cr, x_offset, y_offset,
448 qpix_op, chroma_op);
449
450 qpix_op= qpix_avg;
451 chroma_op= chroma_avg;
452 }
453
454 if(list1){
455 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
456 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
457 dest_y, dest_cb, dest_cr, x_offset, y_offset,
458 qpix_op, chroma_op);
459 }
460 }
461
462 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
463 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
464 int x_offset, int y_offset,
465 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
466 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
467 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
468 int list0, int list1){
469 MpegEncContext * const s = &h->s;
470
471 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
472 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
473 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
474 x_offset += 8*s->mb_x;
475 y_offset += 8*(s->mb_y >> MB_FIELD);
476
477 if(list0 && list1){
478 /* don't optimize for luma-only case, since B-frames usually
479 * use implicit weights => chroma too. */
480 uint8_t *tmp_cb = s->obmc_scratchpad;
481 uint8_t *tmp_cr = s->obmc_scratchpad + 8;
482 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
483 int refn0 = h->ref_cache[0][ scan8[n] ];
484 int refn1 = h->ref_cache[1][ scan8[n] ];
485
486 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
487 dest_y, dest_cb, dest_cr,
488 x_offset, y_offset, qpix_put, chroma_put);
489 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
490 tmp_y, tmp_cb, tmp_cr,
491 x_offset, y_offset, qpix_put, chroma_put);
492
493 if(h->use_weight == 2){
494 int weight0 = h->implicit_weight[refn0][refn1];
495 int weight1 = 64 - weight0;
496 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
497 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
498 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
499 }else{
500 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
501 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
502 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
503 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
504 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
505 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
506 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
507 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
508 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
509 }
510 }else{
511 int list = list1 ? 1 : 0;
512 int refn = h->ref_cache[list][ scan8[n] ];
513 Picture *ref= &h->ref_list[list][refn];
514 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
515 dest_y, dest_cb, dest_cr, x_offset, y_offset,
516 qpix_put, chroma_put);
517
518 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
519 h->luma_weight[list][refn], h->luma_offset[list][refn]);
520 if(h->use_weight_chroma){
521 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
522 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
523 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
524 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
525 }
526 }
527 }
528
529 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
530 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
531 int x_offset, int y_offset,
532 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
533 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
534 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
535 int list0, int list1){
536 if((h->use_weight==2 && list0 && list1
537 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
538 || h->use_weight==1)
539 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
540 x_offset, y_offset, qpix_put, chroma_put,
541 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
542 else
543 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
544 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
545 }
546
547 static inline void prefetch_motion(H264Context *h, int list){
548 /* fetch pixels for estimated mv 4 macroblocks ahead
549 * optimized for 64byte cache lines */
550 MpegEncContext * const s = &h->s;
551 const int refn = h->ref_cache[list][scan8[0]];
552 if(refn >= 0){
553 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
554 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
555 uint8_t **src= h->ref_list[list][refn].data;
556 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
557 s->dsp.prefetch(src[0]+off, s->linesize, 4);
558 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
559 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
560 }
561 }
562
563 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
564 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
565 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
566 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
567 MpegEncContext * const s = &h->s;
568 const int mb_xy= h->mb_xy;
569 const int mb_type= s->current_picture.mb_type[mb_xy];
570
571 assert(IS_INTER(mb_type));
572
573 prefetch_motion(h, 0);
574
575 if(IS_16X16(mb_type)){
576 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
577 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
578 weight_op, weight_avg,
579 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
580 }else if(IS_16X8(mb_type)){
581 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
582 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
583 &weight_op[1], &weight_avg[1],
584 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
585 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
586 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
587 &weight_op[1], &weight_avg[1],
588 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
589 }else if(IS_8X16(mb_type)){
590 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
591 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
592 &weight_op[2], &weight_avg[2],
593 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
594 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
595 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
596 &weight_op[2], &weight_avg[2],
597 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
598 }else{
599 int i;
600
601 assert(IS_8X8(mb_type));
602
603 for(i=0; i<4; i++){
604 const int sub_mb_type= h->sub_mb_type[i];
605 const int n= 4*i;
606 int x_offset= (i&1)<<2;
607 int y_offset= (i&2)<<1;
608
609 if(IS_SUB_8X8(sub_mb_type)){
610 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
611 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
612 &weight_op[3], &weight_avg[3],
613 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
614 }else if(IS_SUB_8X4(sub_mb_type)){
615 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
616 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
617 &weight_op[4], &weight_avg[4],
618 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
619 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
620 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
621 &weight_op[4], &weight_avg[4],
622 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
623 }else if(IS_SUB_4X8(sub_mb_type)){
624 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
625 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
626 &weight_op[5], &weight_avg[5],
627 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
628 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
629 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
630 &weight_op[5], &weight_avg[5],
631 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
632 }else{
633 int j;
634 assert(IS_SUB_4X4(sub_mb_type));
635 for(j=0; j<4; j++){
636 int sub_x_offset= x_offset + 2*(j&1);
637 int sub_y_offset= y_offset + (j&2);
638 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
639 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
640 &weight_op[6], &weight_avg[6],
641 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
642 }
643 }
644 }
645 }
646
647 prefetch_motion(h, 1);
648 }
649
650
651 static void free_tables(H264Context *h){
652 int i;
653 H264Context *hx;
654 av_freep(&h->intra4x4_pred_mode);
655 av_freep(&h->chroma_pred_mode_table);
656 av_freep(&h->cbp_table);
657 av_freep(&h->mvd_table[0]);
658 av_freep(&h->mvd_table[1]);
659 av_freep(&h->direct_table);
660 av_freep(&h->non_zero_count);
661 av_freep(&h->slice_table_base);
662 h->slice_table= NULL;
663 av_freep(&h->list_counts);
664
665 av_freep(&h->mb2b_xy);
666 av_freep(&h->mb2b8_xy);
667
668 for(i = 0; i < MAX_THREADS; i++) {
669 hx = h->thread_context[i];
670 if(!hx) continue;
671 av_freep(&hx->top_borders[1]);
672 av_freep(&hx->top_borders[0]);
673 av_freep(&hx->s.obmc_scratchpad);
674 av_freep(&hx->rbsp_buffer[1]);
675 av_freep(&hx->rbsp_buffer[0]);
676 hx->rbsp_buffer_size[0] = 0;
677 hx->rbsp_buffer_size[1] = 0;
678 if (i) av_freep(&h->thread_context[i]);
679 }
680 }
681
682 static void init_dequant8_coeff_table(H264Context *h){
683 int i,q,x;
684 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
685 h->dequant8_coeff[0] = h->dequant8_buffer[0];
686 h->dequant8_coeff[1] = h->dequant8_buffer[1];
687
688 for(i=0; i<2; i++ ){
689 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
690 h->dequant8_coeff[1] = h->dequant8_buffer[0];
691 break;
692 }
693
694 for(q=0; q<52; q++){
695 int shift = div6[q];
696 int idx = rem6[q];
697 for(x=0; x<64; x++)
698 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
699 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
700 h->pps.scaling_matrix8[i][x]) << shift;
701 }
702 }
703 }
704
705 static void init_dequant4_coeff_table(H264Context *h){
706 int i,j,q,x;
707 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
708 for(i=0; i<6; i++ ){
709 h->dequant4_coeff[i] = h->dequant4_buffer[i];
710 for(j=0; j<i; j++){
711 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
712 h->dequant4_coeff[i] = h->dequant4_buffer[j];
713 break;
714 }
715 }
716 if(j<i)
717 continue;
718
719 for(q=0; q<52; q++){
720 int shift = div6[q] + 2;
721 int idx = rem6[q];
722 for(x=0; x<16; x++)
723 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
724 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
725 h->pps.scaling_matrix4[i][x]) << shift;
726 }
727 }
728 }
729
730 static void init_dequant_tables(H264Context *h){
731 int i,x;
732 init_dequant4_coeff_table(h);
733 if(h->pps.transform_8x8_mode)
734 init_dequant8_coeff_table(h);
735 if(h->sps.transform_bypass){
736 for(i=0; i<6; i++)
737 for(x=0; x<16; x++)
738 h->dequant4_coeff[i][0][x] = 1<<6;
739 if(h->pps.transform_8x8_mode)
740 for(i=0; i<2; i++)
741 for(x=0; x<64; x++)
742 h->dequant8_coeff[i][0][x] = 1<<6;
743 }
744 }
745
746
747 int ff_h264_alloc_tables(H264Context *h){
748 MpegEncContext * const s = &h->s;
749 const int big_mb_num= s->mb_stride * (s->mb_height+1);
750 int x,y;
751
752 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t), fail)
753
754 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count , big_mb_num * 32 * sizeof(uint8_t), fail)
755 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
756 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
757
758 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
759 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t), fail);
760 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t), fail);
761 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 32*big_mb_num * sizeof(uint8_t) , fail);
762 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
763
764 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
765 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
766
767 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy , big_mb_num * sizeof(uint32_t), fail);
768 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b8_xy , big_mb_num * sizeof(uint32_t), fail);
769 for(y=0; y<s->mb_height; y++){
770 for(x=0; x<s->mb_width; x++){
771 const int mb_xy= x + y*s->mb_stride;
772 const int b_xy = 4*x + 4*y*h->b_stride;
773 const int b8_xy= 2*x + 2*y*h->b8_stride;
774
775 h->mb2b_xy [mb_xy]= b_xy;
776 h->mb2b8_xy[mb_xy]= b8_xy;
777 }
778 }
779
780 s->obmc_scratchpad = NULL;
781
782 if(!h->dequant4_coeff[0])
783 init_dequant_tables(h);
784
785 return 0;
786 fail:
787 free_tables(h);
788 return -1;
789 }
790
791 /**
792 * Mimic alloc_tables(), but for every context thread.
793 */
794 static void clone_tables(H264Context *dst, H264Context *src){
795 dst->intra4x4_pred_mode = src->intra4x4_pred_mode;
796 dst->non_zero_count = src->non_zero_count;
797 dst->slice_table = src->slice_table;
798 dst->cbp_table = src->cbp_table;
799 dst->mb2b_xy = src->mb2b_xy;
800 dst->mb2b8_xy = src->mb2b8_xy;
801 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
802 dst->mvd_table[0] = src->mvd_table[0];
803 dst->mvd_table[1] = src->mvd_table[1];
804 dst->direct_table = src->direct_table;
805 dst->list_counts = src->list_counts;
806
807 dst->s.obmc_scratchpad = NULL;
808 ff_h264_pred_init(&dst->hpc, src->s.codec_id);
809 }
810
811 /**
812 * Init context
813 * Allocate buffers which are not shared amongst multiple threads.
814 */
815 static int context_init(H264Context *h){
816 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
817 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
818
819 return 0;
820 fail:
821 return -1; // free_tables will clean up for us
822 }
823
824 static av_cold void common_init(H264Context *h){
825 MpegEncContext * const s = &h->s;
826
827 s->width = s->avctx->width;
828 s->height = s->avctx->height;
829 s->codec_id= s->avctx->codec->id;
830
831 ff_h264_pred_init(&h->hpc, s->codec_id);
832
833 h->dequant_coeff_pps= -1;
834 s->unrestricted_mv=1;
835 s->decode=1; //FIXME
836
837 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
838
839 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
840 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
841 }
842
843 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
844 H264Context *h= avctx->priv_data;
845 MpegEncContext * const s = &h->s;
846
847 MPV_decode_defaults(s);
848
849 s->avctx = avctx;
850 common_init(h);
851
852 s->out_format = FMT_H264;
853 s->workaround_bugs= avctx->workaround_bugs;
854
855 // set defaults
856 // s->decode_mb= ff_h263_decode_mb;
857 s->quarter_sample = 1;
858 if(!avctx->has_b_frames)
859 s->low_delay= 1;
860
861 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
862
863 ff_h264_decode_init_vlc();
864
865 if(avctx->extradata_size > 0 && avctx->extradata &&
866 *(char *)avctx->extradata == 1){
867 h->is_avc = 1;
868 h->got_avcC = 0;
869 } else {
870 h->is_avc = 0;
871 }
872
873 h->thread_context[0] = h;
874 h->outputed_poc = INT_MIN;
875 h->prev_poc_msb= 1<<16;
876 ff_h264_reset_sei(h);
877 if(avctx->codec_id == CODEC_ID_H264){
878 if(avctx->ticks_per_frame == 1){
879 s->avctx->time_base.den *=2;
880 }
881 avctx->ticks_per_frame = 2;
882 }
883 return 0;
884 }
885
886 int ff_h264_frame_start(H264Context *h){
887 MpegEncContext * const s = &h->s;
888 int i;
889
890 if(MPV_frame_start(s, s->avctx) < 0)
891 return -1;
892 ff_er_frame_start(s);
893 /*
894 * MPV_frame_start uses pict_type to derive key_frame.
895 * This is incorrect for H.264; IDR markings must be used.
896 * Zero here; IDR markings per slice in frame or fields are ORed in later.
897 * See decode_nal_units().
898 */
899 s->current_picture_ptr->key_frame= 0;
900 s->current_picture_ptr->mmco_reset= 0;
901
902 assert(s->linesize && s->uvlinesize);
903
904 for(i=0; i<16; i++){
905 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
906 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
907 }
908 for(i=0; i<4; i++){
909 h->block_offset[16+i]=
910 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
911 h->block_offset[24+16+i]=
912 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
913 }
914
915 /* can't be in alloc_tables because linesize isn't known there.
916 * FIXME: redo bipred weight to not require extra buffer? */
917 for(i = 0; i < s->avctx->thread_count; i++)
918 if(!h->thread_context[i]->s.obmc_scratchpad)
919 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
920
921 /* some macroblocks will be accessed before they're available */
922 if(FRAME_MBAFF || s->avctx->thread_count > 1)
923 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
924
925 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
926
927 // We mark the current picture as non-reference after allocating it, so
928 // that if we break out due to an error it can be released automatically
929 // in the next MPV_frame_start().
930 // SVQ3 as well as most other codecs have only last/next/current and thus
931 // get released even with set reference, besides SVQ3 and others do not
932 // mark frames as reference later "naturally".
933 if(s->codec_id != CODEC_ID_SVQ3)
934 s->current_picture_ptr->reference= 0;
935
936 s->current_picture_ptr->field_poc[0]=
937 s->current_picture_ptr->field_poc[1]= INT_MAX;
938 assert(s->current_picture_ptr->long_ref==0);
939
940 return 0;
941 }
942
943 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
944 MpegEncContext * const s = &h->s;
945 uint8_t *top_border;
946 int top_idx = 1;
947
948 src_y -= linesize;
949 src_cb -= uvlinesize;
950 src_cr -= uvlinesize;
951
952 if(!simple && FRAME_MBAFF){
953 if(s->mb_y&1){
954 if(!MB_MBAFF){
955 top_border = h->top_borders[0][s->mb_x];
956 AV_COPY128(top_border, src_y + 15*linesize);
957 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
958 AV_COPY64(top_border+16, src_cb+7*uvlinesize);
959 AV_COPY64(top_border+24, src_cr+7*uvlinesize);
960 }
961 }
962 }else if(MB_MBAFF){
963 top_idx = 0;
964 }else
965 return;
966 }
967
968 top_border = h->top_borders[top_idx][s->mb_x];
969 // There are two lines saved, the line above the the top macroblock of a pair,
970 // and the line above the bottom macroblock
971 AV_COPY128(top_border, src_y + 16*linesize);
972
973 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
974 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
975 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
976 }
977 }
978
979 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
980 MpegEncContext * const s = &h->s;
981 int temp8, i;
982 uint64_t temp64;
983 int deblock_left;
984 int deblock_top;
985 int mb_xy;
986 int top_idx = 1;
987 uint8_t *top_border_m1;
988 uint8_t *top_border;
989
990 if(!simple && FRAME_MBAFF){
991 if(s->mb_y&1){
992 if(!MB_MBAFF)
993 return;
994 }else{
995 top_idx = MB_MBAFF ? 0 : 1;
996 }
997 }
998
999 if(h->deblocking_filter == 2) {
1000 mb_xy = h->mb_xy;
1001 deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
1002 deblock_top = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
1003 } else {
1004 deblock_left = (s->mb_x > 0);
1005 deblock_top = (s->mb_y > !!MB_FIELD);
1006 }
1007
1008 src_y -= linesize + 1;
1009 src_cb -= uvlinesize + 1;
1010 src_cr -= uvlinesize + 1;
1011
1012 top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1013 top_border = h->top_borders[top_idx][s->mb_x];
1014
1015 #define XCHG(a,b,xchg)\
1016 if (xchg) AV_SWAP64(b,a);\
1017 else AV_COPY64(b,a);
1018
1019 if(deblock_top){
1020 if(deblock_left){
1021 XCHG(top_border_m1+8, src_y -7, 1);
1022 }
1023 XCHG(top_border+0, src_y +1, xchg);
1024 XCHG(top_border+8, src_y +9, 1);
1025 if(s->mb_x+1 < s->mb_width){
1026 XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1027 }
1028 }
1029
1030 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1031 if(deblock_top){
1032 if(deblock_left){
1033 XCHG(top_border_m1+16, src_cb -7, 1);
1034 XCHG(top_border_m1+24, src_cr -7, 1);
1035 }
1036 XCHG(top_border+16, src_cb+1, 1);
1037 XCHG(top_border+24, src_cr+1, 1);
1038 }
1039 }
1040 }
1041
1042 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1043 MpegEncContext * const s = &h->s;
1044 const int mb_x= s->mb_x;
1045 const int mb_y= s->mb_y;
1046 const int mb_xy= h->mb_xy;
1047 const int mb_type= s->current_picture.mb_type[mb_xy];
1048 uint8_t *dest_y, *dest_cb, *dest_cr;
1049 int linesize, uvlinesize /*dct_offset*/;
1050 int i;
1051 int *block_offset = &h->block_offset[0];
1052 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1053 /* is_h264 should always be true if SVQ3 is disabled. */
1054 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1055 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1056 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1057
1058 dest_y = s->current_picture.data[0] + (mb_x + mb_y * s->linesize ) * 16;
1059 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1060 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1061
1062 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1063 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1064
1065 h->list_counts[mb_xy]= h->list_count;
1066
1067 if (!simple && MB_FIELD) {
1068 linesize = h->mb_linesize = s->linesize * 2;
1069 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1070 block_offset = &h->block_offset[24];
1071 if(mb_y&1){ //FIXME move out of this function?
1072 dest_y -= s->linesize*15;
1073 dest_cb-= s->uvlinesize*7;
1074 dest_cr-= s->uvlinesize*7;
1075 }
1076 if(FRAME_MBAFF) {
1077 int list;
1078 for(list=0; list<h->list_count; list++){
1079 if(!USES_LIST(mb_type, list))
1080 continue;
1081 if(IS_16X16(mb_type)){
1082 int8_t *ref = &h->ref_cache[list][scan8[0]];
1083 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1084 }else{
1085 for(i=0; i<16; i+=4){
1086 int ref = h->ref_cache[list][scan8[i]];
1087 if(ref >= 0)
1088 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1089 }
1090 }
1091 }
1092 }
1093 } else {
1094 linesize = h->mb_linesize = s->linesize;
1095 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1096 // dct_offset = s->linesize * 16;
1097 }
1098
1099 if (!simple && IS_INTRA_PCM(mb_type)) {
1100 for (i=0; i<16; i++) {
1101 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
1102 }
1103 for (i=0; i<8; i++) {
1104 memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
1105 memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
1106 }
1107 } else {
1108 if(IS_INTRA(mb_type)){
1109 if(h->deblocking_filter)
1110 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1111
1112 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1113 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1114 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1115 }
1116
1117 if(IS_INTRA4x4(mb_type)){
1118 if(simple || !s->encoding){
1119 if(IS_8x8DCT(mb_type)){
1120 if(transform_bypass){
1121 idct_dc_add =
1122 idct_add = s->dsp.add_pixels8;
1123 }else{
1124 idct_dc_add = s->dsp.h264_idct8_dc_add;
1125 idct_add = s->dsp.h264_idct8_add;
1126 }
1127 for(i=0; i<16; i+=4){
1128 uint8_t * const ptr= dest_y + block_offset[i];
1129 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1130 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1131 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1132 }else{
1133 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1134 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1135 (h->topright_samples_available<<i)&0x4000, linesize);
1136 if(nnz){
1137 if(nnz == 1 && h->mb[i*16])
1138 idct_dc_add(ptr, h->mb + i*16, linesize);
1139 else
1140 idct_add (ptr, h->mb + i*16, linesize);
1141 }
1142 }
1143 }
1144 }else{
1145 if(transform_bypass){
1146 idct_dc_add =
1147 idct_add = s->dsp.add_pixels4;
1148 }else{
1149 idct_dc_add = s->dsp.h264_idct_dc_add;
1150 idct_add = s->dsp.h264_idct_add;
1151 }
1152 for(i=0; i<16; i++){
1153 uint8_t * const ptr= dest_y + block_offset[i];
1154 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1155
1156 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1157 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1158 }else{
1159 uint8_t *topright;
1160 int nnz, tr;
1161 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1162 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1163 assert(mb_y || linesize <= block_offset[i]);
1164 if(!topright_avail){
1165 tr= ptr[3 - linesize]*0x01010101;
1166 topright= (uint8_t*) &tr;
1167 }else
1168 topright= ptr + 4 - linesize;
1169 }else
1170 topright= NULL;
1171
1172 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1173 nnz = h->non_zero_count_cache[ scan8[i] ];
1174 if(nnz){
1175 if(is_h264){
1176 if(nnz == 1 && h->mb[i*16])
1177 idct_dc_add(ptr, h->mb + i*16, linesize);
1178 else
1179 idct_add (ptr, h->mb + i*16, linesize);
1180 }else
1181 ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1182 }
1183 }
1184 }
1185 }
1186 }
1187 }else{
1188 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1189 if(is_h264){
1190 if(!transform_bypass)
1191 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1192 }else
1193 ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1194 }
1195 if(h->deblocking_filter)
1196 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1197 }else if(is_h264){
1198 hl_motion(h, dest_y, dest_cb, dest_cr,
1199 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1200 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1201 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1202 }
1203
1204
1205 if(!IS_INTRA4x4(mb_type)){
1206 if(is_h264){
1207 if(IS_INTRA16x16(mb_type)){
1208 if(transform_bypass){
1209 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1210 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1211 }else{
1212 for(i=0; i<16; i++){
1213 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1214 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1215 }
1216 }
1217 }else{
1218 s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1219 }
1220 }else if(h->cbp&15){
1221 if(transform_bypass){
1222 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1223 idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1224 for(i=0; i<16; i+=di){
1225 if(h->non_zero_count_cache[ scan8[i] ]){
1226 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1227 }
1228 }
1229 }else{
1230 if(IS_8x8DCT(mb_type)){
1231 s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1232 }else{
1233 s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1234 }
1235 }
1236 }
1237 }else{
1238 for(i=0; i<16; i++){
1239 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1240 uint8_t * const ptr= dest_y + block_offset[i];
1241 ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1242 }
1243 }
1244 }
1245 }
1246
1247 if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1248 uint8_t *dest[2] = {dest_cb, dest_cr};
1249 if(transform_bypass){
1250 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1251 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1252 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1253 }else{
1254 idct_add = s->dsp.add_pixels4;
1255 for(i=16; i<16+8; i++){
1256 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1257 idct_add (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1258 }
1259 }
1260 }else{
1261 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp[0], h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1262 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp[1], h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1263 if(is_h264){
1264 idct_add = s->dsp.h264_idct_add;
1265 idct_dc_add = s->dsp.h264_idct_dc_add;
1266 for(i=16; i<16+8; i++){
1267 if(h->non_zero_count_cache[ scan8[i] ])
1268 idct_add (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1269 else if(h->mb[i*16])
1270 idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1271 }
1272 }else{
1273 for(i=16; i<16+8; i++){
1274 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1275 uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1276 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1277 }
1278 }
1279 }
1280 }
1281 }
1282 }
1283 if(h->cbp || IS_INTRA(mb_type))
1284 s->dsp.clear_blocks(h->mb);
1285
1286 if(h->deblocking_filter && 0) {
1287 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
1288 fill_filter_caches(h, mb_type); //FIXME don't fill stuff which isn't used by filter_mb
1289 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
1290 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
1291 if (!simple && FRAME_MBAFF) {
1292 ff_h264_filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1293 } else {
1294 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1295 }
1296 }
1297 }
1298
1299 /**
1300 * Process a macroblock; this case avoids checks for expensive uncommon cases.
1301 */
1302 static void hl_decode_mb_simple(H264Context *h){
1303 hl_decode_mb_internal(h, 1);
1304 }
1305
1306 /**
1307 * Process a macroblock; this handles edge cases, such as interlacing.
1308 */
1309 static void av_noinline hl_decode_mb_complex(H264Context *h){
1310 hl_decode_mb_internal(h, 0);
1311 }
1312
1313 void ff_h264_hl_decode_mb(H264Context *h){
1314 MpegEncContext * const s = &h->s;
1315 const int mb_xy= h->mb_xy;
1316 const int mb_type= s->current_picture.mb_type[mb_xy];
1317 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1318
1319 if (is_complex)
1320 hl_decode_mb_complex(h);
1321 else hl_decode_mb_simple(h);
1322 }
1323
1324 static int pred_weight_table(H264Context *h){
1325 MpegEncContext * const s = &h->s;
1326 int list, i;
1327 int luma_def, chroma_def;
1328
1329 h->use_weight= 0;
1330 h->use_weight_chroma= 0;
1331 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1332 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1333 luma_def = 1<<h->luma_log2_weight_denom;
1334 chroma_def = 1<<h->chroma_log2_weight_denom;
1335
1336 for(list=0; list<2; list++){
1337 h->luma_weight_flag[list] = 0;
1338 h->chroma_weight_flag[list] = 0;
1339 for(i=0; i<h->ref_count[list]; i++){
1340 int luma_weight_flag, chroma_weight_flag;
1341
1342 luma_weight_flag= get_bits1(&s->gb);
1343 if(luma_weight_flag){
1344 h->luma_weight[list][i]= get_se_golomb(&s->gb);
1345 h->luma_offset[list][i]= get_se_golomb(&s->gb);
1346 if( h->luma_weight[list][i] != luma_def
1347 || h->luma_offset[list][i] != 0) {
1348 h->use_weight= 1;
1349 h->luma_weight_flag[list]= 1;
1350 }
1351 }else{
1352 h->luma_weight[list][i]= luma_def;
1353 h->luma_offset[list][i]= 0;
1354 }
1355
1356 if(CHROMA){
1357 chroma_weight_flag= get_bits1(&s->gb);
1358 if(chroma_weight_flag){
1359 int j;
1360 for(j=0; j<2; j++){
1361 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1362 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1363 if( h->chroma_weight[list][i][j] != chroma_def
1364 || h->chroma_offset[list][i][j] != 0) {
1365 h->use_weight_chroma= 1;
1366 h->chroma_weight_flag[list]= 1;
1367 }
1368 }
1369 }else{
1370 int j;
1371 for(j=0; j<2; j++){
1372 h->chroma_weight[list][i][j]= chroma_def;
1373 h->chroma_offset[list][i][j]= 0;
1374 }
1375 }
1376 }
1377 }
1378 if(h->slice_type_nos != FF_B_TYPE) break;
1379 }
1380 h->use_weight= h->use_weight || h->use_weight_chroma;
1381 return 0;
1382 }
1383
1384 static void implicit_weight_table(H264Context *h){
1385 MpegEncContext * const s = &h->s;
1386 int ref0, ref1, i;
1387 int cur_poc = s->current_picture_ptr->poc;
1388
1389 for (i = 0; i < 2; i++) {
1390 h->luma_weight_flag[i] = 0;
1391 h->chroma_weight_flag[i] = 0;
1392 }
1393
1394 if( h->ref_count[0] == 1 && h->ref_count[1] == 1
1395 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1396 h->use_weight= 0;
1397 h->use_weight_chroma= 0;
1398 return;
1399 }
1400
1401 h->use_weight= 2;
1402 h->use_weight_chroma= 2;
1403 h->luma_log2_weight_denom= 5;
1404 h->chroma_log2_weight_denom= 5;
1405
1406 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1407 int poc0 = h->ref_list[0][ref0].poc;
1408 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1409 int poc1 = h->ref_list[1][ref1].poc;
1410 int td = av_clip(poc1 - poc0, -128, 127);
1411 if(td){
1412 int tb = av_clip(cur_poc - poc0, -128, 127);
1413 int tx = (16384 + (FFABS(td) >> 1)) / td;
1414 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1415 if(dist_scale_factor < -64 || dist_scale_factor > 128)
1416 h->implicit_weight[ref0][ref1] = 32;
1417 else
1418 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1419 }else
1420 h->implicit_weight[ref0][ref1] = 32;
1421 }
1422 }
1423 }
1424
1425 /**
1426 * instantaneous decoder refresh.
1427 */
1428 static void idr(H264Context *h){
1429 ff_h264_remove_all_refs(h);
1430 h->prev_frame_num= 0;
1431 h->prev_frame_num_offset= 0;
1432 h->prev_poc_msb=
1433 h->prev_poc_lsb= 0;
1434 }
1435
1436 /* forget old pics after a seek */
1437 static void flush_dpb(AVCodecContext *avctx){
1438 H264Context *h= avctx->priv_data;
1439 int i;
1440 for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1441 if(h->delayed_pic[i])
1442 h->delayed_pic[i]->reference= 0;
1443 h->delayed_pic[i]= NULL;
1444 }
1445 h->outputed_poc= INT_MIN;
1446 h->prev_interlaced_frame = 1;
1447 idr(h);
1448 if(h->s.current_picture_ptr)
1449 h->s.current_picture_ptr->reference= 0;
1450 h->s.first_field= 0;
1451 ff_h264_reset_sei(h);
1452 ff_mpeg_flush(avctx);
1453 }
1454
1455 static int init_poc(H264Context *h){
1456 MpegEncContext * const s = &h->s;
1457 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1458 int field_poc[2];
1459 Picture *cur = s->current_picture_ptr;
1460
1461 h->frame_num_offset= h->prev_frame_num_offset;
1462 if(h->frame_num < h->prev_frame_num)
1463 h->frame_num_offset += max_frame_num;
1464
1465 if(h->sps.poc_type==0){
1466 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1467
1468 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1469 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1470 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1471 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1472 else
1473 h->poc_msb = h->prev_poc_msb;
1474 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1475 field_poc[0] =
1476 field_poc[1] = h->poc_msb + h->poc_lsb;
1477 if(s->picture_structure == PICT_FRAME)
1478 field_poc[1] += h->delta_poc_bottom;
1479 }else if(h->sps.poc_type==1){
1480 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1481 int i;
1482
1483 if(h->sps.poc_cycle_length != 0)
1484 abs_frame_num = h->frame_num_offset + h->frame_num;
1485 else
1486 abs_frame_num = 0;
1487
1488 if(h->nal_ref_idc==0 && abs_frame_num > 0)
1489 abs_frame_num--;
1490
1491 expected_delta_per_poc_cycle = 0;
1492 for(i=0; i < h->sps.poc_cycle_length; i++)
1493 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1494
1495 if(abs_frame_num > 0){
1496 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1497 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1498
1499 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1500 for(i = 0; i <= frame_num_in_poc_cycle; i++)
1501 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1502 } else
1503 expectedpoc = 0;
1504
1505 if(h->nal_ref_idc == 0)
1506 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1507
1508 field_poc[0] = expectedpoc + h->delta_poc[0];
1509 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1510
1511 if(s->picture_structure == PICT_FRAME)
1512 field_poc[1] += h->delta_poc[1];
1513 }else{
1514 int poc= 2*(h->frame_num_offset + h->frame_num);
1515
1516 if(!h->nal_ref_idc)
1517 poc--;
1518
1519 field_poc[0]= poc;
1520 field_poc[1]= poc;
1521 }
1522
1523 if(s->picture_structure != PICT_BOTTOM_FIELD)
1524 s->current_picture_ptr->field_poc[0]= field_poc[0];
1525 if(s->picture_structure != PICT_TOP_FIELD)
1526 s->current_picture_ptr->field_poc[1]= field_poc[1];
1527 cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1528
1529 return 0;
1530 }
1531
1532
1533 /**
1534 * initialize scan tables
1535 */
1536 static void init_scan_tables(H264Context *h){
1537 MpegEncContext * const s = &h->s;
1538 int i;
1539 if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1540 memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1541 memcpy(h-> field_scan, field_scan, 16*sizeof(uint8_t));
1542 }else{
1543 for(i=0; i<16; i++){
1544 #define T(x) (x>>2) | ((x<<2) & 0xF)
1545 h->zigzag_scan[i] = T(zigzag_scan[i]);
1546 h-> field_scan[i] = T( field_scan[i]);
1547 #undef T
1548 }
1549 }
1550 if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1551 memcpy(h->zigzag_scan8x8, ff_zigzag_direct, 64*sizeof(uint8_t));
1552 memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1553 memcpy(h->field_scan8x8, field_scan8x8, 64*sizeof(uint8_t));
1554 memcpy(h->field_scan8x8_cavlc, field_scan8x8_cavlc, 64*sizeof(uint8_t));
1555 }else{
1556 for(i=0; i<64; i++){
1557 #define T(x) (x>>3) | ((x&7)<<3)
1558 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
1559 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1560 h->field_scan8x8[i] = T(field_scan8x8[i]);
1561 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
1562 #undef T
1563 }
1564 }
1565 if(h->sps.transform_bypass){ //FIXME same ugly
1566 h->zigzag_scan_q0 = zigzag_scan;
1567 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
1568 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1569 h->field_scan_q0 = field_scan;
1570 h->field_scan8x8_q0 = field_scan8x8;
1571 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
1572 }else{
1573 h->zigzag_scan_q0 = h->zigzag_scan;
1574 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
1575 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1576 h->field_scan_q0 = h->field_scan;
1577 h->field_scan8x8_q0 = h->field_scan8x8;
1578 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
1579 }
1580 }
1581
1582 static void field_end(H264Context *h){
1583 MpegEncContext * const s = &h->s;
1584 AVCodecContext * const avctx= s->avctx;
1585 s->mb_y= 0;
1586
1587 s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1588 s->current_picture_ptr->pict_type= s->pict_type;
1589
1590 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1591 ff_vdpau_h264_set_reference_frames(s);
1592
1593 if(!s->dropable) {
1594 ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1595 h->prev_poc_msb= h->poc_msb;
1596 h->prev_poc_lsb= h->poc_lsb;
1597 }
1598 h->prev_frame_num_offset= h->frame_num_offset;
1599 h->prev_frame_num= h->frame_num;
1600
1601 if (avctx->hwaccel) {
1602 if (avctx->hwaccel->end_frame(avctx) < 0)
1603 av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1604 }
1605
1606 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1607 ff_vdpau_h264_picture_complete(s);
1608
1609 /*
1610 * FIXME: Error handling code does not seem to support interlaced
1611 * when slices span multiple rows
1612 * The ff_er_add_slice calls don't work right for bottom
1613 * fields; they cause massive erroneous error concealing
1614 * Error marking covers both fields (top and bottom).
1615 * This causes a mismatched s->error_count
1616 * and a bad error table. Further, the error count goes to
1617 * INT_MAX when called for bottom field, because mb_y is
1618 * past end by one (callers fault) and resync_mb_y != 0
1619 * causes problems for the first MB line, too.
1620 */
1621 if (!FIELD_PICTURE)
1622 ff_er_frame_end(s);
1623
1624 MPV_frame_end(s);
1625
1626 h->current_slice=0;
1627 }
1628
1629 /**
1630 * Replicates H264 "master" context to thread contexts.
1631 */
1632 static void clone_slice(H264Context *dst, H264Context *src)
1633 {
1634 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
1635 dst->s.current_picture_ptr = src->s.current_picture_ptr;
1636 dst->s.current_picture = src->s.current_picture;
1637 dst->s.linesize = src->s.linesize;
1638 dst->s.uvlinesize = src->s.uvlinesize;
1639 dst->s.first_field = src->s.first_field;
1640
1641 dst->prev_poc_msb = src->prev_poc_msb;
1642 dst->prev_poc_lsb = src->prev_poc_lsb;
1643 dst->prev_frame_num_offset = src->prev_frame_num_offset;
1644 dst->prev_frame_num = src->prev_frame_num;
1645 dst->short_ref_count = src->short_ref_count;
1646
1647 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
1648 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
1649 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1650 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
1651
1652 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
1653 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
1654 }
1655
1656 /**
1657 * decodes a slice header.
1658 * This will also call MPV_common_init() and frame_start() as needed.
1659 *
1660 * @param h h264context
1661 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1662 *
1663 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1664 */
1665 static int decode_slice_header(H264Context *h, H264Context *h0){
1666 MpegEncContext * const s = &h->s;
1667 MpegEncContext * const s0 = &h0->s;
1668 unsigned int first_mb_in_slice;
1669 unsigned int pps_id;
1670 int num_ref_idx_active_override_flag;
1671 unsigned int slice_type, tmp, i, j;
1672 int default_ref_list_done = 0;
1673 int last_pic_structure;
1674
1675 s->dropable= h->nal_ref_idc == 0;
1676
1677 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1678 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1679 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1680 }else{
1681 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1682 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1683 }
1684
1685 first_mb_in_slice= get_ue_golomb(&s->gb);
1686
1687 if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1688 if(h0->current_slice && FIELD_PICTURE){
1689 field_end(h);
1690 }
1691
1692 h0->current_slice = 0;
1693 if (!s0->first_field)
1694 s->current_picture_ptr= NULL;
1695 }
1696
1697 slice_type= get_ue_golomb_31(&s->gb);
1698 if(slice_type > 9){
1699 av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
1700 return -1;
1701 }
1702 if(slice_type > 4){
1703 slice_type -= 5;
1704 h->slice_type_fixed=1;
1705 }else
1706 h->slice_type_fixed=0;
1707
1708 slice_type= golomb_to_pict_type[ slice_type ];
1709 if (slice_type == FF_I_TYPE
1710 || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1711 default_ref_list_done = 1;
1712 }
1713 h->slice_type= slice_type;
1714 h->slice_type_nos= slice_type & 3;
1715
1716 s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1717 if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
1718 av_log(h->s.avctx, AV_LOG_ERROR,
1719 "B picture before any references, skipping\n");
1720 return -1;
1721 }
1722
1723 pps_id= get_ue_golomb(&s->gb);
1724 if(pps_id>=MAX_PPS_COUNT){
1725 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1726 return -1;
1727 }
1728 if(!h0->pps_buffers[pps_id]) {
1729 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1730 return -1;
1731 }
1732 h->pps= *h0->pps_buffers[pps_id];
1733
1734 if(!h0->sps_buffers[h->pps.sps_id]) {
1735 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1736 return -1;
1737 }
1738 h->sps = *h0->sps_buffers[h->pps.sps_id];
1739
1740 if(h == h0 && h->dequant_coeff_pps != pps_id){
1741 h->dequant_coeff_pps = pps_id;
1742 init_dequant_tables(h);
1743 }
1744
1745 s->mb_width= h->sps.mb_width;
1746 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1747
1748 h->b_stride= s->mb_width*4;
1749 h->b8_stride= s->mb_width*2;
1750
1751 s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1752 if(h->sps.frame_mbs_only_flag)
1753 s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1754 else
1755 s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1756
1757 if (s->context_initialized
1758 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
1759 if(h != h0)
1760 return -1; // width / height changed during parallelized decoding
1761 free_tables(h);
1762 flush_dpb(s->avctx);
1763 MPV_common_end(s);
1764 }
1765 if (!s->context_initialized) {
1766 if(h != h0)
1767 return -1; // we cant (re-)initialize context during parallel decoding
1768
1769 avcodec_set_dimensions(s->avctx, s->width, s->height);
1770 s->avctx->sample_aspect_ratio= h->sps.sar;
1771 if(!s->avctx->sample_aspect_ratio.den)
1772 s->avctx->sample_aspect_ratio.den = 1;
1773
1774 if(h->sps.video_signal_type_present_flag){
1775 s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1776 if(h->sps.colour_description_present_flag){
1777 s->avctx->color_primaries = h->sps.color_primaries;
1778 s->avctx->color_trc = h->sps.color_trc;
1779 s->avctx->colorspace = h->sps.colorspace;
1780 }
1781 }
1782
1783 if(h->sps.timing_info_present_flag){
1784 s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
1785 if(h->x264_build > 0 && h->x264_build < 44)
1786 s->avctx->time_base.den *= 2;
1787 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1788 s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
1789 }
1790 s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1791 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1792
1793 if (MPV_common_init(s) < 0)
1794 return -1;
1795 s->first_field = 0;
1796 h->prev_interlaced_frame = 1;
1797
1798 init_scan_tables(h);
1799 ff_h264_alloc_tables(h);
1800
1801 for(i = 1; i < s->avctx->thread_count; i++) {
1802 H264Context *c;
1803 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1804 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1805 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1806 c->sps = h->sps;
1807 c->pps = h->pps;
1808 init_scan_tables(c);
1809 clone_tables(c, h);
1810 }
1811
1812 for(i = 0; i < s->avctx->thread_count; i++)
1813 if(context_init(h->thread_context[i]) < 0)
1814 return -1;
1815 }
1816
1817 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1818
1819 h->mb_mbaff = 0;
1820 h->mb_aff_frame = 0;
1821 last_pic_structure = s0->picture_structure;
1822 if(h->sps.frame_mbs_only_flag){
1823 s->picture_structure= PICT_FRAME;
1824 }else{
1825 if(get_bits1(&s->gb)) { //field_pic_flag
1826 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1827 } else {
1828 s->picture_structure= PICT_FRAME;
1829 h->mb_aff_frame = h->sps.mb_aff;
1830 }
1831 }
1832 h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1833
1834 if(h0->current_slice == 0){
1835 while(h->frame_num != h->prev_frame_num &&
1836 h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1837 av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1838 if (ff_h264_frame_start(h) < 0)
1839 return -1;
1840 h->prev_frame_num++;
1841 h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1842 s->current_picture_ptr->frame_num= h->prev_frame_num;
1843 ff_h264_execute_ref_pic_marking(h, NULL, 0);
1844 }
1845
1846 /* See if we have a decoded first field looking for a pair... */
1847 if (s0->first_field) {
1848 assert(s0->current_picture_ptr);
1849 assert(s0->current_picture_ptr->data[0]);
1850 assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1851
1852 /* figure out if we have a complementary field pair */
1853 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1854 /*
1855 * Previous field is unmatched. Don't display it, but let it
1856 * remain for reference if marked as such.
1857 */
1858 s0->current_picture_ptr = NULL;
1859 s0->first_field = FIELD_PICTURE;
1860
1861 } else {
1862 if (h->nal_ref_idc &&
1863 s0->current_picture_ptr->reference &&
1864 s0->current_picture_ptr->frame_num != h->frame_num) {
1865 /*
1866 * This and previous field were reference, but had
1867 * different frame_nums. Consider this field first in
1868 * pair. Throw away previous field except for reference
1869 * purposes.
1870 */
1871 s0->first_field = 1;
1872 s0->current_picture_ptr = NULL;
1873
1874 } else {
1875 /* Second field in complementary pair */
1876 s0->first_field = 0;
1877 }
1878 }
1879
1880 } else {
1881 /* Frame or first field in a potentially complementary pair */
1882 assert(!s0->current_picture_ptr);
1883 s0->first_field = FIELD_PICTURE;
1884 }
1885
1886 if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1887 s0->first_field = 0;
1888 return -1;
1889 }
1890 }
1891 if(h != h0)
1892 clone_slice(h, h0);
1893
1894 s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1895
1896 assert(s->mb_num == s->mb_width * s->mb_height);
1897 if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1898 first_mb_in_slice >= s->mb_num){
1899 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1900 return -1;
1901 }
1902 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1903 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1904 if (s->picture_structure == PICT_BOTTOM_FIELD)
1905 s->resync_mb_y = s->mb_y = s->mb_y + 1;
1906 assert(s->mb_y < s->mb_height);
1907
1908 if(s->picture_structure==PICT_FRAME){
1909 h->curr_pic_num= h->frame_num;
1910 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1911 }else{
1912 h->curr_pic_num= 2*h->frame_num + 1;
1913 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1914 }
1915
1916 if(h->nal_unit_type == NAL_IDR_SLICE){
1917 get_ue_golomb(&s->gb); /* idr_pic_id */
1918 }
1919
1920 if(h->sps.poc_type==0){
1921 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1922
1923 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1924 h->delta_poc_bottom= get_se_golomb(&s->gb);
1925 }
1926 }
1927
1928 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1929 h->delta_poc[0]= get_se_golomb(&s->gb);
1930
1931 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1932 h->delta_poc[1]= get_se_golomb(&s->gb);
1933 }
1934
1935 init_poc(h);
1936
1937 if(h->pps.redundant_pic_cnt_present){
1938 h->redundant_pic_count= get_ue_golomb(&s->gb);
1939 }
1940
1941 //set defaults, might be overridden a few lines later
1942 h->ref_count[0]= h->pps.ref_count[0];
1943 h->ref_count[1]= h->pps.ref_count[1];
1944
1945 if(h->slice_type_nos != FF_I_TYPE){
1946 if(h->slice_type_nos == FF_B_TYPE){
1947 h->direct_spatial_mv_pred= get_bits1(&s->gb);
1948 }
1949 num_ref_idx_active_override_flag= get_bits1(&s->gb);
1950
1951 if(num_ref_idx_active_override_flag){
1952 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1953 if(h->slice_type_nos==FF_B_TYPE)
1954 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1955
1956 if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1957 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1958 h->ref_count[0]= h->ref_count[1]= 1;
1959 return -1;
1960 }
1961 }
1962 if(h->slice_type_nos == FF_B_TYPE)
1963 h->list_count= 2;
1964 else
1965 h->list_count= 1;
1966 }else
1967 h->list_count= 0;
1968
1969 if(!default_ref_list_done){
1970 ff_h264_fill_default_ref_list(h);
1971 }
1972
1973 if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
1974 return -1;
1975
1976 if(h->slice_type_nos!=FF_I_TYPE){
1977 s->last_picture_ptr= &h->ref_list[0][0];
1978 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
1979 }
1980 if(h->slice_type_nos==FF_B_TYPE){
1981 s->next_picture_ptr= &h->ref_list[1][0];
1982 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
1983 }
1984
1985 if( (h->pps.weighted_pred && h->slice_type_nos == FF_P_TYPE )
1986 || (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
1987 pred_weight_table(h);
1988 else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
1989 implicit_weight_table(h);
1990 else {
1991 h->use_weight = 0;
1992 for (i = 0; i < 2; i++) {
1993 h->luma_weight_flag[i] = 0;
1994 h->chroma_weight_flag[i] = 0;
1995 }
1996 }
1997
1998 if(h->nal_ref_idc)
1999 ff_h264_decode_ref_pic_marking(h0, &s->gb);
2000
2001 if(FRAME_MBAFF)
2002 ff_h264_fill_mbaff_ref_list(h);
2003
2004 if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2005 ff_h264_direct_dist_scale_factor(h);
2006 ff_h264_direct_ref_list_init(h);
2007
2008 if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2009 tmp = get_ue_golomb_31(&s->gb);
2010 if(tmp > 2){
2011 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2012 return -1;
2013 }
2014 h->cabac_init_idc= tmp;
2015 }
2016
2017 h->last_qscale_diff = 0;
2018 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2019 if(tmp>51){
2020 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2021 return -1;
2022 }
2023 s->qscale= tmp;
2024 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2025 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2026 //FIXME qscale / qp ... stuff
2027 if(h->slice_type == FF_SP_TYPE){
2028 get_bits1(&s->gb); /* sp_for_switch_flag */
2029 }
2030 if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2031 get_se_golomb(&s->gb); /* slice_qs_delta */
2032 }
2033
2034 h->deblocking_filter = 1;
2035 h->slice_alpha_c0_offset = 52;
2036 h->slice_beta_offset = 52;
2037 if( h->pps.deblocking_filter_parameters_present ) {
2038 tmp= get_ue_golomb_31(&s->gb);
2039 if(tmp > 2){
2040 av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2041 return -1;
2042 }
2043 h->deblocking_filter= tmp;
2044 if(h->deblocking_filter < 2)
2045 h->deblocking_filter^= 1; // 1<->0
2046
2047 if( h->deblocking_filter ) {
2048 h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2049 h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
2050 if( h->slice_alpha_c0_offset > 104U
2051 || h->slice_beta_offset > 104U){
2052 av_log(s->avctx, AV_LOG_ERROR, "deblocking filter parameters %d %d out of range\n", h->slice_alpha_c0_offset, h->slice_beta_offset);
2053 return -1;
2054 }
2055 }
2056 }
2057
2058 if( s->avctx->skip_loop_filter >= AVDISCARD_ALL
2059 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2060 ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR && h->slice_type_nos == FF_B_TYPE)
2061 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2062 h->deblocking_filter= 0;
2063
2064 if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2065 if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2066 /* Cheat slightly for speed:
2067 Do not bother to deblock across slices. */
2068 h->deblocking_filter = 2;
2069 } else {
2070 h0->max_contexts = 1;
2071 if(!h0->single_decode_warning) {
2072 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2073 h0->single_decode_warning = 1;
2074 }
2075 if(h != h0)
2076 return 1; // deblocking switched inside frame
2077 }
2078 }
2079 h->qp_thresh= 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
2080
2081 #if 0 //FMO
2082 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2083 slice_group_change_cycle= get_bits(&s->gb, ?);
2084 #endif
2085
2086 h0->last_slice_type = slice_type;
2087 h->slice_num = ++h0->current_slice;
2088 if(h->slice_num >= MAX_SLICES){
2089 av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2090 }
2091
2092 for(j=0; j<2; j++){
2093 int id_list[16];
2094 int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2095 for(i=0; i<16; i++){
2096 id_list[i]= 60;
2097 if(h->ref_list[j][i].data[0]){
2098 int k;
2099 uint8_t *base= h->ref_list[j][i].base[0];
2100 for(k=0; k<h->short_ref_count; k++)
2101 if(h->short_ref[k]->base[0] == base){
2102 id_list[i]= k;
2103 break;
2104 }
2105 for(k=0; k<h->long_ref_count; k++)
2106 if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2107 id_list[i]= h->short_ref_count + k;
2108 break;
2109 }
2110 }
2111 }
2112
2113 ref2frm[0]=
2114 ref2frm[1]= -1;
2115 for(i=0; i<16; i++)
2116 ref2frm[i+2]= 4*id_list[i]
2117 +(h->ref_list[j][i].reference&3);
2118 ref2frm[18+0]=
2119 ref2frm[18+1]= -1;
2120 for(i=16; i<48; i++)
2121 ref2frm[i+4]= 4*id_list[(i-16)>>1]
2122 +(h->ref_list[j][i].reference&3);
2123 }
2124
2125 h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2126 h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2127
2128 s->avctx->refs= h->sps.ref_frame_count;
2129
2130 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2131 av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
2132 h->slice_num,
2133 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2134 first_mb_in_slice,
2135 av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2136 pps_id, h->frame_num,
2137 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2138 h->ref_count[0], h->ref_count[1],
2139 s->qscale,
2140 h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2141 h->use_weight,
2142 h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2143 h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2144 );
2145 }
2146
2147 return 0;
2148 }
2149
2150 int ff_h264_get_slice_type(const H264Context *h)
2151 {
2152 switch (h->slice_type) {
2153 case FF_P_TYPE: return 0;
2154 case FF_B_TYPE: return 1;
2155 case FF_I_TYPE: return 2;
2156 case FF_SP_TYPE: return 3;
2157 case FF_SI_TYPE: return 4;
2158 default: return -1;
2159 }
2160 }
2161
2162 static void loop_filter(H264Context *h){
2163 MpegEncContext * const s = &h->s;
2164 uint8_t *dest_y, *dest_cb, *dest_cr;
2165 int linesize, uvlinesize, mb_x, mb_y;
2166 const int end_mb_y= s->mb_y + FRAME_MBAFF;
2167 const int old_slice_type= h->slice_type;
2168
2169 if(h->deblocking_filter) {
2170 for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2171 for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2172 int list, mb_xy, mb_type;
2173 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2174 h->slice_num= h->slice_table[mb_xy];
2175 mb_type= s->current_picture.mb_type[mb_xy];
2176 h->list_count= h->list_counts[mb_xy];
2177
2178 if(FRAME_MBAFF)
2179 h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2180
2181 s->mb_x= mb_x;
2182 s->mb_y= mb_y;
2183 dest_y = s->current_picture.data[0] + (mb_x + mb_y * s->linesize ) * 16;
2184 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2185 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2186 //FIXME simplify above
2187
2188 if (MB_FIELD) {
2189 linesize = h->mb_linesize = s->linesize * 2;
2190 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2191 if(mb_y&1){ //FIXME move out of this function?
2192 dest_y -= s->linesize*15;
2193 dest_cb-= s->uvlinesize*7;
2194 dest_cr-= s->uvlinesize*7;
2195 }
2196 } else {
2197 linesize = h->mb_linesize = s->linesize;
2198 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2199 }
2200 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2201 if(fill_filter_caches(h, mb_type))
2202 continue;
2203 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2204 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2205
2206 if (FRAME_MBAFF) {
2207 ff_h264_filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2208 } else {
2209 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2210 }
2211 }
2212 }
2213 }
2214 h->slice_type= old_slice_type;
2215 s->mb_x= 0;
2216 s->mb_y= end_mb_y - FRAME_MBAFF;
2217 }
2218
2219 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2220 H264Context *h = *(void**)arg;
2221 MpegEncContext * const s = &h->s;
2222 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2223
2224 s->mb_skip_run= -1;
2225
2226 h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2227 (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2228
2229 if( h->pps.cabac ) {
2230 /* realign */
2231 align_get_bits( &s->gb );
2232
2233 /* init cabac */
2234 ff_init_cabac_states( &h->cabac);
2235 ff_init_cabac_decoder( &h->cabac,
2236 s->gb.buffer + get_bits_count(&s->gb)/8,
2237 (get_bits_left(&s->gb) + 7)/8);
2238
2239 ff_h264_init_cabac_states(h);
2240
2241 for(;;){
2242 //START_TIMER
2243 int ret = ff_h264_decode_mb_cabac(h);
2244 int eos;
2245 //STOP_TIMER("decode_mb_cabac")
2246
2247 if(ret>=0) ff_h264_hl_decode_mb(h);
2248
2249 if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2250 s->mb_y++;
2251
2252 ret = ff_h264_decode_mb_cabac(h);
2253
2254 if(ret>=0) ff_h264_hl_decode_mb(h);
2255 s->mb_y--;
2256 }
2257 eos = get_cabac_terminate( &h->cabac );
2258
2259 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2260 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%td)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);
2261 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);
2262 return -1;
2263 }
2264
2265 if( ++s->mb_x >= s->mb_width ) {
2266 s->mb_x = 0;
2267 loop_filter(h);
2268 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2269 ++s->mb_y;
2270 if(FIELD_OR_MBAFF_PICTURE) {
2271 ++s->mb_y;
2272 }
2273 }
2274
2275 if( eos || s->mb_y >= s->mb_height ) {
2276 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2277 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);
2278 return 0;
2279 }
2280 }
2281
2282 } else {
2283 for(;;){
2284 int ret = ff_h264_decode_mb_cavlc(h);
2285
2286 if(ret>=0) ff_h264_hl_decode_mb(h);
2287
2288 if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2289 s->mb_y++;
2290 ret = ff_h264_decode_mb_cavlc(h);
2291
2292 if(ret>=0) ff_h264_hl_decode_mb(h);
2293 s->mb_y--;
2294 }
2295
2296 if(ret<0){
2297 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2298 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);
2299
2300 return -1;
2301 }
2302
2303 if(++s->mb_x >= s->mb_width){
2304 s->mb_x=0;
2305 loop_filter(h);
2306 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2307 ++s->mb_y;
2308 if(FIELD_OR_MBAFF_PICTURE) {
2309 ++s->mb_y;
2310 }
2311 if(s->mb_y >= s->mb_height){
2312 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2313
2314 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2315 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);
2316
2317 return 0;
2318 }else{
2319 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);
2320
2321 return -1;
2322 }
2323 }
2324 }
2325
2326 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2327 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2328 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2329 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);
2330
2331 return 0;
2332 }else{
2333 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);
2334
2335 return -1;
2336 }
2337 }
2338 }
2339 }
2340
2341 #if 0
2342 for(;s->mb_y < s->mb_height; s->mb_y++){
2343 for(;s->mb_x < s->mb_width; s->mb_x++){
2344 int ret= decode_mb(h);
2345
2346 ff_h264_hl_decode_mb(h);
2347
2348 if(ret<0){
2349 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2350 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);
2351
2352 return -1;
2353 }
2354
2355 if(++s->mb_x >= s->mb_width){
2356 s->mb_x=0;
2357 if(++s->mb_y >= s->mb_height){
2358 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2359 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);
2360
2361 return 0;
2362 }else{
2363 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);
2364
2365 return -1;
2366 }
2367 }
2368 }
2369
2370 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2371 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2372 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);
2373
2374 return 0;
2375 }else{
2376 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);
2377
2378 return -1;
2379 }
2380 }
2381 }
2382 s->mb_x=0;
2383 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2384 }
2385 #endif
2386 return -1; //not reached
2387 }
2388
2389 /**
2390 * Call decode_slice() for each context.
2391 *
2392 * @param h h264 master context
2393 * @param context_count number of contexts to execute
2394 */
2395 static void execute_decode_slices(H264Context *h, int context_count){
2396 MpegEncContext * const s = &h->s;
2397 AVCodecContext * const avctx= s->avctx;
2398 H264Context *hx;
2399 int i;
2400
2401 if (s->avctx->hwaccel)
2402 return;
2403 if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2404 return;
2405 if(context_count == 1) {
2406 decode_slice(avctx, &h);
2407 } else {
2408 for(i = 1; i < context_count; i++) {
2409 hx = h->thread_context[i];
2410 hx->s.error_recognition = avctx->error_recognition;
2411 hx->s.error_count = 0;
2412 }
2413
2414 avctx->execute(avctx, (void *)decode_slice,
2415 h->thread_context, NULL, context_count, sizeof(void*));
2416
2417 /* pull back stuff from slices to master context */
2418 hx = h->thread_context[context_count - 1];
2419 s->mb_x = hx->s.mb_x;
2420 s->mb_y = hx->s.mb_y;
2421 s->dropable = hx->s.dropable;
2422 s->picture_structure = hx->s.picture_structure;
2423 for(i = 1; i < context_count; i++)
2424 h->s.error_count += h->thread_context[i]->s.error_count;
2425 }
2426 }
2427
2428
2429 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2430 MpegEncContext * const s = &h->s;
2431 AVCodecContext * const avctx= s->avctx;
2432 int buf_index=0;
2433 H264Context *hx; ///< thread context
2434 int context_count = 0;
2435 int next_avc= h->is_avc ? 0 : buf_size;
2436
2437 h->max_contexts = avctx->thread_count;
2438 #if 0
2439 int i;
2440 for(i=0; i<50; i++){
2441 av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2442 }
2443 #endif
2444 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2445 h->current_slice = 0;
2446 if (!s->first_field)
2447 s->current_picture_ptr= NULL;
2448 ff_h264_reset_sei(h);
2449 }
2450
2451 for(;;){
2452 int consumed;
2453 int dst_length;
2454 int bit_length;
2455 const uint8_t *ptr;
2456 int i, nalsize = 0;
2457 int err;
2458
2459 if(buf_index >= next_avc) {
2460 if(buf_index >= buf_size) break;
2461 nalsize = 0;
2462 for(i = 0; i < h->nal_length_size; i++)
2463 nalsize = (nalsize << 8) | buf[buf_index++];
2464 if(nalsize <= 1 || nalsize > buf_size - buf_index){
2465 if(nalsize == 1){
2466 buf_index++;
2467 continue;
2468 }else{
2469 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2470 break;
2471 }
2472 }
2473 next_avc= buf_index + nalsize;
2474 } else {
2475 // start code prefix search
2476 for(; buf_index + 3 < next_avc; buf_index++){
2477 // This should always succeed in the first iteration.
2478 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2479 break;
2480 }
2481
2482 if(buf_index+3 >= buf_size) break;
2483
2484 buf_index+=3;
2485 if(buf_index >= next_avc) continue;
2486 }
2487
2488 hx = h->thread_context[context_count];
2489
2490 ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2491 if (ptr==NULL || dst_length < 0){
2492 return -1;
2493 }
2494 while(ptr[dst_length - 1] == 0 && dst_length > 0)
2495 dst_length--;
2496 bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2497
2498 if(s->avctx->debug&FF_DEBUG_STARTCODE){
2499 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", hx->nal_unit_type, buf_index, buf_size, dst_length);
2500 }
2501
2502 if (h->is_avc && (nalsize != consumed) && nalsize){
2503 av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2504 }
2505
2506 buf_index += consumed;
2507
2508 if( (s->hurry_up == 1 && h->nal_ref_idc == 0) //FIXME do not discard SEI id
2509 ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2510 continue;
2511
2512 again:
2513 err = 0;
2514 switch(hx->nal_unit_type){
2515 case NAL_IDR_SLICE:
2516 if (h->nal_unit_type != NAL_IDR_SLICE) {
2517 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2518 return -1;
2519 }
2520 idr(h); //FIXME ensure we don't loose some frames if there is reordering
2521 case NAL_SLICE:
2522 init_get_bits(&hx->s.gb, ptr, bit_length);
2523 hx->intra_gb_ptr=
2524 hx->inter_gb_ptr= &hx->s.gb;
2525 hx->s.data_partitioning = 0;
2526
2527 if((err = decode_slice_header(hx, h)))
2528 break;
2529
2530 if (s->avctx->hwaccel && h->current_slice == 1) {
2531 if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2532 return -1;
2533 }
2534
2535 s->current_picture_ptr->key_frame |=
2536 (hx->nal_unit_type == NAL_IDR_SLICE) ||
2537 (h->sei_recovery_frame_cnt >= 0);
2538 if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2539 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2540 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=FF_B_TYPE)
2541 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2542 && avctx->skip_frame < AVDISCARD_ALL){
2543 if(avctx->hwaccel) {
2544 if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2545 return -1;
2546 }else
2547 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2548 static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2549 ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2550 ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2551 }else
2552 context_count++;
2553 }
2554 break;
2555 case NAL_DPA:
2556 init_get_bits(&hx->s.gb, ptr, bit_length);
2557 hx->intra_gb_ptr=
2558 hx->inter_gb_ptr= NULL;
2559
2560 if ((err = decode_slice_header(hx, h)) < 0)
2561 break;
2562
2563 hx->s.data_partitioning = 1;
2564
2565 break;
2566 case NAL_DPB:
2567 init_get_bits(&hx->intra_gb, ptr, bit_length);
2568 hx->intra_gb_ptr= &hx->intra_gb;
2569 break;
2570 case NAL_DPC:
2571 init_get_bits(&hx->inter_gb, ptr, bit_length);
2572 hx->inter_gb_ptr= &hx->inter_gb;
2573
2574 if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2575 && s->context_initialized
2576 && s->hurry_up < 5
2577 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2578 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=FF_B_TYPE)
2579 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2580 && avctx->skip_frame < AVDISCARD_ALL)
2581 context_count++;
2582 break;
2583 case NAL_SEI:
2584 init_get_bits(&s->gb, ptr, bit_length);
2585 ff_h264_decode_sei(h);
2586 break;
2587 case NAL_SPS:
2588 init_get_bits(&s->gb, ptr, bit_length);
2589 ff_h264_decode_seq_parameter_set(h);
2590
2591 if(s->flags& CODEC_FLAG_LOW_DELAY)
2592 s->low_delay=1;
2593
2594 if(avctx->has_b_frames < 2)
2595 avctx->has_b_frames= !s->low_delay;
2596 break;
2597 case NAL_PPS:
2598 init_get_bits(&s->gb, ptr, bit_length);
2599
2600 ff_h264_decode_picture_parameter_set(h, bit_length);
2601
2602 break;
2603 case NAL_AUD:
2604 case NAL_END_SEQUENCE:
2605 case NAL_END_STREAM:
2606 case NAL_FILLER_DATA:
2607 case NAL_SPS_EXT:
2608 case NAL_AUXILIARY_SLICE:
2609 break;
2610 default:
2611 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2612 }
2613
2614 if(context_count == h->max_contexts) {
2615 execute_decode_slices(h, context_count);
2616 context_count = 0;
2617 }
2618
2619 if (err < 0)
2620 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2621 else if(err == 1) {
2622 /* Slice could not be decoded in parallel mode, copy down
2623 * NAL unit stuff to context 0 and restart. Note that
2624 * rbsp_buffer is not transferred, but since we no longer
2625 * run in parallel mode this should not be an issue. */
2626 h->nal_unit_type = hx->nal_unit_type;
2627 h->nal_ref_idc = hx->nal_ref_idc;
2628 hx = h;
2629 goto again;
2630 }
2631 }
2632 if(context_count)
2633 execute_decode_slices(h, context_count);
2634 return buf_index;
2635 }
2636
2637 /**
2638 * returns the number of bytes consumed for building the current frame
2639 */
2640 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2641 if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2642 if(pos+10>buf_size) pos=buf_size; // oops ;)
2643
2644 return pos;
2645 }
2646
2647 static int decode_frame(AVCodecContext *avctx,
2648 void *data, int *data_size,
2649 AVPacket *avpkt)
2650 {
2651 const uint8_t *buf = avpkt->data;
2652 int buf_size = avpkt->size;
2653 H264Context *h = avctx->priv_data;
2654 MpegEncContext *s = &h->s;
2655 AVFrame *pict = data;
2656 int buf_index;
2657
2658 s->flags= avctx->flags;
2659 s->flags2= avctx->flags2;
2660
2661 /* end of stream, output what is still in the buffers */
2662 if (buf_size == 0) {
2663 Picture *out;
2664 int i, out_idx;
2665
2666 //FIXME factorize this with the output code below
2667 out = h->delayed_pic[0];
2668 out_idx = 0;
2669 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2670 if(h->delayed_pic[i]->poc < out->poc){
2671 out = h->delayed_pic[i];
2672 out_idx = i;
2673 }
2674
2675 for(i=out_idx; h->delayed_pic[i]; i++)
2676 h->delayed_pic[i] = h->delayed_pic[i+1];
2677
2678 if(out){
2679 *data_size = sizeof(AVFrame);
2680 *pict= *(AVFrame*)out;
2681 }
2682
2683 return 0;
2684 }
2685
2686 if(h->is_avc && !h->got_avcC) {
2687 int i, cnt, nalsize;
2688 unsigned char *p = avctx->extradata;
2689 if(avctx->extradata_size < 7) {
2690 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
2691 return -1;
2692 }
2693 if(*p != 1) {
2694 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
2695 return -1;
2696 }
2697 /* sps and pps in the avcC always have length coded with 2 bytes,
2698 so put a fake nal_length_size = 2 while parsing them */
2699 h->nal_length_size = 2;
2700 // Decode sps from avcC
2701 cnt = *(p+5) & 0x1f; // Number of sps
2702 p += 6;
2703 for (i = 0; i < cnt; i++) {
2704 nalsize = AV_RB16(p) + 2;
2705 if(decode_nal_units(h, p, nalsize) < 0) {
2706 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
2707 return -1;
2708 }
2709 p += nalsize;
2710 }
2711 // Decode pps from avcC
2712 cnt = *(p++); // Number of pps
2713 for (i = 0; i < cnt; i++) {
2714 nalsize = AV_RB16(p) + 2;
2715 if(decode_nal_units(h, p, nalsize) != nalsize) {
2716 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
2717 return -1;
2718 }
2719 p += nalsize;
2720 }
2721 // Now store right nal length size, that will be use to parse all other nals
2722 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
2723 // Do not reparse avcC
2724 h->got_avcC = 1;
2725 }
2726
2727 if(!h->got_avcC && !h->is_avc && s->avctx->extradata_size){
2728 if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
2729 return -1;
2730 h->got_avcC = 1;
2731 }
2732
2733 buf_index=decode_nal_units(h, buf, buf_size);
2734 if(buf_index < 0)
2735 return -1;
2736
2737 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2738 if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2739 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2740 return -1;
2741 }
2742
2743 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2744 Picture *out = s->current_picture_ptr;
2745 Picture *cur = s->current_picture_ptr;
2746 int i, pics, out_of_order, out_idx;
2747
2748 field_end(h);
2749
2750 if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2751 /* Wait for second field. */
2752 *data_size = 0;
2753
2754 } else {
2755 cur->interlaced_frame = 0;
2756 cur->repeat_pict = 0;
2757
2758 /* Signal interlacing information externally. */
2759 /* Prioritize picture timing SEI information over used decoding process if it exists. */
2760
2761 if(h->sps.pic_struct_present_flag){
2762 switch (h->sei_pic_struct)
2763 {
2764 case SEI_PIC_STRUCT_FRAME:
2765 break;
2766 case SEI_PIC_STRUCT_TOP_FIELD:
2767 case SEI_PIC_STRUCT_BOTTOM_FIELD:
2768 cur->interlaced_frame = 1;
2769 break;
2770 case SEI_PIC_STRUCT_TOP_BOTTOM:
2771 case SEI_PIC_STRUCT_BOTTOM_TOP:
2772 if (FIELD_OR_MBAFF_PICTURE)
2773 cur->interlaced_frame = 1;
2774 else
2775 // try to flag soft telecine progressive
2776 cur->interlaced_frame = h->prev_interlaced_frame;
2777 break;
2778 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2779 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2780 // Signal the possibility of telecined film externally (pic_struct 5,6)
2781 // From these hints, let the applications decide if they apply deinterlacing.
2782 cur->repeat_pict = 1;
2783 break;