acd7179cc0ca16ac3a8b575ab783d7407053494d
[libav.git] / libavcodec / h264.c
1 /*
2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This file is part of Libav.
6 *
7 * Libav 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 * Libav 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 Libav; 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
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28 #include "libavutil/imgutils.h"
29 #include "internal.h"
30 #include "dsputil.h"
31 #include "avcodec.h"
32 #include "mpegvideo.h"
33 #include "h264.h"
34 #include "h264data.h"
35 #include "h264_mvpred.h"
36 #include "golomb.h"
37 #include "mathops.h"
38 #include "rectangle.h"
39 #include "thread.h"
40 #include "vdpau_internal.h"
41 #include "libavutil/avassert.h"
42
43 #include "cabac.h"
44
45 //#undef NDEBUG
46 #include <assert.h>
47
48 static const uint8_t rem6[QP_MAX_NUM+1]={
49 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, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
50 };
51
52 static const uint8_t div6[QP_MAX_NUM+1]={
53 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, 8, 8, 9, 9, 9, 9, 9, 9,10,10,10,10,
54 };
55
56 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
57 PIX_FMT_DXVA2_VLD,
58 PIX_FMT_VAAPI_VLD,
59 PIX_FMT_VDA_VLD,
60 PIX_FMT_YUVJ420P,
61 PIX_FMT_NONE
62 };
63
64 /**
65 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
66 */
67 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
68 MpegEncContext * const s = &h->s;
69 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
70 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
71 int i;
72
73 if(!(h->top_samples_available&0x8000)){
74 for(i=0; i<4; i++){
75 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
76 if(status<0){
77 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);
78 return -1;
79 } else if(status){
80 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
81 }
82 }
83 }
84
85 if((h->left_samples_available&0x8888)!=0x8888){
86 static const int mask[4]={0x8000,0x2000,0x80,0x20};
87 for(i=0; i<4; i++){
88 if(!(h->left_samples_available&mask[i])){
89 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
90 if(status<0){
91 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);
92 return -1;
93 } else if(status){
94 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
95 }
96 }
97 }
98 }
99
100 return 0;
101 } //FIXME cleanup like ff_h264_check_intra_pred_mode
102
103 /**
104 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
105 */
106 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
107 MpegEncContext * const s = &h->s;
108 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
109 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
110
111 if(mode > 6U) {
112 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);
113 return -1;
114 }
115
116 if(!(h->top_samples_available&0x8000)){
117 mode= top[ mode ];
118 if(mode<0){
119 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);
120 return -1;
121 }
122 }
123
124 if((h->left_samples_available&0x8080) != 0x8080){
125 mode= left[ mode ];
126 if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
127 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
128 }
129 if(mode<0){
130 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);
131 return -1;
132 }
133 }
134
135 return mode;
136 }
137
138 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
139 int i, si, di;
140 uint8_t *dst;
141 int bufidx;
142
143 // src[0]&0x80; //forbidden bit
144 h->nal_ref_idc= src[0]>>5;
145 h->nal_unit_type= src[0]&0x1F;
146
147 src++; length--;
148
149 #if HAVE_FAST_UNALIGNED
150 # if HAVE_FAST_64BIT
151 # define RS 7
152 for(i=0; i+1<length; i+=9){
153 if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
154 # else
155 # define RS 3
156 for(i=0; i+1<length; i+=5){
157 if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
158 # endif
159 continue;
160 if(i>0 && !src[i]) i--;
161 while(src[i]) i++;
162 #else
163 # define RS 0
164 for(i=0; i+1<length; i+=2){
165 if(src[i]) continue;
166 if(i>0 && src[i-1]==0) i--;
167 #endif
168 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
169 if(src[i+2]!=3){
170 /* startcode, so we must be past the end */
171 length=i;
172 }
173 break;
174 }
175 i-= RS;
176 }
177
178 if(i>=length-1){ //no escaped 0
179 *dst_length= length;
180 *consumed= length+1; //+1 for the header
181 return src;
182 }
183
184 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
185 av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
186 dst= h->rbsp_buffer[bufidx];
187
188 if (dst == NULL){
189 return NULL;
190 }
191
192 //printf("decoding esc\n");
193 memcpy(dst, src, i);
194 si=di=i;
195 while(si+2<length){
196 //remove escapes (very rare 1:2^22)
197 if(src[si+2]>3){
198 dst[di++]= src[si++];
199 dst[di++]= src[si++];
200 }else if(src[si]==0 && src[si+1]==0){
201 if(src[si+2]==3){ //escape
202 dst[di++]= 0;
203 dst[di++]= 0;
204 si+=3;
205 continue;
206 }else //next start code
207 goto nsc;
208 }
209
210 dst[di++]= src[si++];
211 }
212 while(si<length)
213 dst[di++]= src[si++];
214 nsc:
215
216 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
217
218 *dst_length= di;
219 *consumed= si + 1;//+1 for the header
220 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
221 return dst;
222 }
223
224 /**
225 * Identify the exact end of the bitstream
226 * @return the length of the trailing, or 0 if damaged
227 */
228 static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
229 int v= *src;
230 int r;
231
232 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
233
234 for(r=1; r<9; r++){
235 if(v&1) return r;
236 v>>=1;
237 }
238 return 0;
239 }
240
241 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height,
242 int y_offset, int list){
243 int raw_my= h->mv_cache[list][ scan8[n] ][1];
244 int filter_height= (raw_my&3) ? 2 : 0;
245 int full_my= (raw_my>>2) + y_offset;
246 int top = full_my - filter_height, bottom = full_my + height + filter_height;
247
248 return FFMAX(abs(top), bottom);
249 }
250
251 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height,
252 int y_offset, int list0, int list1, int *nrefs){
253 MpegEncContext * const s = &h->s;
254 int my;
255
256 y_offset += 16*(s->mb_y >> MB_FIELD);
257
258 if(list0){
259 int ref_n = h->ref_cache[0][ scan8[n] ];
260 Picture *ref= &h->ref_list[0][ref_n];
261
262 // Error resilience puts the current picture in the ref list.
263 // Don't try to wait on these as it will cause a deadlock.
264 // Fields can wait on each other, though.
265 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
266 (ref->f.reference & 3) != s->picture_structure) {
267 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
268 if (refs[0][ref_n] < 0) nrefs[0] += 1;
269 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
270 }
271 }
272
273 if(list1){
274 int ref_n = h->ref_cache[1][ scan8[n] ];
275 Picture *ref= &h->ref_list[1][ref_n];
276
277 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
278 (ref->f.reference & 3) != s->picture_structure) {
279 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
280 if (refs[1][ref_n] < 0) nrefs[1] += 1;
281 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
282 }
283 }
284 }
285
286 /**
287 * Wait until all reference frames are available for MC operations.
288 *
289 * @param h the H264 context
290 */
291 static void await_references(H264Context *h){
292 MpegEncContext * const s = &h->s;
293 const int mb_xy= h->mb_xy;
294 const int mb_type = s->current_picture.f.mb_type[mb_xy];
295 int refs[2][48];
296 int nrefs[2] = {0};
297 int ref, list;
298
299 memset(refs, -1, sizeof(refs));
300
301 if(IS_16X16(mb_type)){
302 get_lowest_part_y(h, refs, 0, 16, 0,
303 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
304 }else if(IS_16X8(mb_type)){
305 get_lowest_part_y(h, refs, 0, 8, 0,
306 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
307 get_lowest_part_y(h, refs, 8, 8, 8,
308 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
309 }else if(IS_8X16(mb_type)){
310 get_lowest_part_y(h, refs, 0, 16, 0,
311 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
312 get_lowest_part_y(h, refs, 4, 16, 0,
313 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
314 }else{
315 int i;
316
317 assert(IS_8X8(mb_type));
318
319 for(i=0; i<4; i++){
320 const int sub_mb_type= h->sub_mb_type[i];
321 const int n= 4*i;
322 int y_offset= (i&2)<<2;
323
324 if(IS_SUB_8X8(sub_mb_type)){
325 get_lowest_part_y(h, refs, n , 8, y_offset,
326 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
327 }else if(IS_SUB_8X4(sub_mb_type)){
328 get_lowest_part_y(h, refs, n , 4, y_offset,
329 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
330 get_lowest_part_y(h, refs, n+2, 4, y_offset+4,
331 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
332 }else if(IS_SUB_4X8(sub_mb_type)){
333 get_lowest_part_y(h, refs, n , 8, y_offset,
334 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
335 get_lowest_part_y(h, refs, n+1, 8, y_offset,
336 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
337 }else{
338 int j;
339 assert(IS_SUB_4X4(sub_mb_type));
340 for(j=0; j<4; j++){
341 int sub_y_offset= y_offset + 2*(j&2);
342 get_lowest_part_y(h, refs, n+j, 4, sub_y_offset,
343 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
344 }
345 }
346 }
347 }
348
349 for(list=h->list_count-1; list>=0; list--){
350 for(ref=0; ref<48 && nrefs[list]; ref++){
351 int row = refs[list][ref];
352 if(row >= 0){
353 Picture *ref_pic = &h->ref_list[list][ref];
354 int ref_field = ref_pic->f.reference - 1;
355 int ref_field_picture = ref_pic->field_picture;
356 int pic_height = 16*s->mb_height >> ref_field_picture;
357
358 row <<= MB_MBAFF;
359 nrefs[list]--;
360
361 if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields
362 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) - !(row&1), pic_height-1), 1);
363 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) , pic_height-1), 0);
364 }else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame
365 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row*2 + ref_field , pic_height-1), 0);
366 }else if(FIELD_PICTURE){
367 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), ref_field);
368 }else{
369 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), 0);
370 }
371 }
372 }
373 }
374 }
375
376 #if 0
377 /**
378 * DCT transforms the 16 dc values.
379 * @param qp quantization parameter ??? FIXME
380 */
381 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
382 // const int qmul= dequant_coeff[qp][0];
383 int i;
384 int temp[16]; //FIXME check if this is a good idea
385 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
386 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
387
388 for(i=0; i<4; i++){
389 const int offset= y_offset[i];
390 const int z0= block[offset+stride*0] + block[offset+stride*4];
391 const int z1= block[offset+stride*0] - block[offset+stride*4];
392 const int z2= block[offset+stride*1] - block[offset+stride*5];
393 const int z3= block[offset+stride*1] + block[offset+stride*5];
394
395 temp[4*i+0]= z0+z3;
396 temp[4*i+1]= z1+z2;
397 temp[4*i+2]= z1-z2;
398 temp[4*i+3]= z0-z3;
399 }
400
401 for(i=0; i<4; i++){
402 const int offset= x_offset[i];
403 const int z0= temp[4*0+i] + temp[4*2+i];
404 const int z1= temp[4*0+i] - temp[4*2+i];
405 const int z2= temp[4*1+i] - temp[4*3+i];
406 const int z3= temp[4*1+i] + temp[4*3+i];
407
408 block[stride*0 +offset]= (z0 + z3)>>1;
409 block[stride*2 +offset]= (z1 + z2)>>1;
410 block[stride*8 +offset]= (z1 - z2)>>1;
411 block[stride*10+offset]= (z0 - z3)>>1;
412 }
413 }
414 #endif
415
416 #undef xStride
417 #undef stride
418
419 #if 0
420 static void chroma_dc_dct_c(DCTELEM *block){
421 const int stride= 16*2;
422 const int xStride= 16;
423 int a,b,c,d,e;
424
425 a= block[stride*0 + xStride*0];
426 b= block[stride*0 + xStride*1];
427 c= block[stride*1 + xStride*0];
428 d= block[stride*1 + xStride*1];
429
430 e= a-b;
431 a= a+b;
432 b= c-d;
433 c= c+d;
434
435 block[stride*0 + xStride*0]= (a+c);
436 block[stride*0 + xStride*1]= (e+b);
437 block[stride*1 + xStride*0]= (a-c);
438 block[stride*1 + xStride*1]= (e-b);
439 }
440 #endif
441
442 static av_always_inline void
443 mc_dir_part(H264Context *h, Picture *pic, int n, int square,
444 int height, int delta, int list,
445 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
446 int src_x_offset, int src_y_offset,
447 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
448 int pixel_shift, int chroma_idc)
449 {
450 MpegEncContext * const s = &h->s;
451 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
452 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
453 const int luma_xy= (mx&3) + ((my&3)<<2);
454 int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
455 uint8_t * src_y = pic->f.data[0] + offset;
456 uint8_t * src_cb, * src_cr;
457 int extra_width= h->emu_edge_width;
458 int extra_height= h->emu_edge_height;
459 int emu=0;
460 const int full_mx= mx>>2;
461 const int full_my= my>>2;
462 const int pic_width = 16*s->mb_width;
463 const int pic_height = 16*s->mb_height >> MB_FIELD;
464 int ysh;
465
466 if(mx&7) extra_width -= 3;
467 if(my&7) extra_height -= 3;
468
469 if( full_mx < 0-extra_width
470 || full_my < 0-extra_height
471 || full_mx + 16/*FIXME*/ > pic_width + extra_width
472 || full_my + 16/*FIXME*/ > pic_height + extra_height){
473 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
474 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
475 src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
476 emu=1;
477 }
478
479 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
480 if(!square){
481 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
482 }
483
484 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
485
486 if(chroma_idc == 3 /* yuv444 */){
487 src_cb = pic->f.data[1] + offset;
488 if(emu){
489 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
490 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
491 src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
492 }
493 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
494 if(!square){
495 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
496 }
497
498 src_cr = pic->f.data[2] + offset;
499 if(emu){
500 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
501 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
502 src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
503 }
504 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
505 if(!square){
506 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
507 }
508 return;
509 }
510
511 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
512 if(chroma_idc == 1 /* yuv420 */ && MB_FIELD){
513 // chroma offset when predicting from a field of opposite parity
514 my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
515 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
516 }
517
518 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
519 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
520
521 if(emu){
522 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
523 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
524 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
525 src_cb= s->edge_emu_buffer;
526 }
527 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
528 mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
529
530 if(emu){
531 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
532 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
533 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
534 src_cr= s->edge_emu_buffer;
535 }
536 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
537 mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
538 }
539
540 static av_always_inline void
541 mc_part_std(H264Context *h, int n, int square, int height, int delta,
542 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
543 int x_offset, int y_offset,
544 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
545 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
546 int list0, int list1, int pixel_shift, int chroma_idc)
547 {
548 MpegEncContext * const s = &h->s;
549 qpel_mc_func *qpix_op= qpix_put;
550 h264_chroma_mc_func chroma_op= chroma_put;
551
552 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
553 if (chroma_idc == 3 /* yuv444 */) {
554 dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
555 dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
556 } else if (chroma_idc == 2 /* yuv422 */) {
557 dest_cb += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
558 dest_cr += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
559 } else /* yuv420 */ {
560 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
561 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
562 }
563 x_offset += 8*s->mb_x;
564 y_offset += 8*(s->mb_y >> MB_FIELD);
565
566 if(list0){
567 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
568 mc_dir_part(h, ref, n, square, height, delta, 0,
569 dest_y, dest_cb, dest_cr, x_offset, y_offset,
570 qpix_op, chroma_op, pixel_shift, chroma_idc);
571
572 qpix_op= qpix_avg;
573 chroma_op= chroma_avg;
574 }
575
576 if(list1){
577 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
578 mc_dir_part(h, ref, n, square, height, delta, 1,
579 dest_y, dest_cb, dest_cr, x_offset, y_offset,
580 qpix_op, chroma_op, pixel_shift, chroma_idc);
581 }
582 }
583
584 static av_always_inline void
585 mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
586 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
587 int x_offset, int y_offset,
588 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
589 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
590 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
591 int list0, int list1, int pixel_shift, int chroma_idc){
592 MpegEncContext * const s = &h->s;
593 int chroma_height;
594
595 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
596 if (chroma_idc == 3 /* yuv444 */) {
597 chroma_height = height;
598 chroma_weight_avg = luma_weight_avg;
599 chroma_weight_op = luma_weight_op;
600 dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
601 dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
602 } else if (chroma_idc == 2 /* yuv422 */) {
603 chroma_height = height;
604 dest_cb += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
605 dest_cr += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
606 } else /* yuv420 */ {
607 chroma_height = height >> 1;
608 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
609 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
610 }
611 x_offset += 8*s->mb_x;
612 y_offset += 8*(s->mb_y >> MB_FIELD);
613
614 if(list0 && list1){
615 /* don't optimize for luma-only case, since B-frames usually
616 * use implicit weights => chroma too. */
617 uint8_t *tmp_cb = s->obmc_scratchpad;
618 uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
619 uint8_t *tmp_y = s->obmc_scratchpad + 16*h->mb_uvlinesize;
620 int refn0 = h->ref_cache[0][ scan8[n] ];
621 int refn1 = h->ref_cache[1][ scan8[n] ];
622
623 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
624 dest_y, dest_cb, dest_cr,
625 x_offset, y_offset, qpix_put, chroma_put,
626 pixel_shift, chroma_idc);
627 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
628 tmp_y, tmp_cb, tmp_cr,
629 x_offset, y_offset, qpix_put, chroma_put,
630 pixel_shift, chroma_idc);
631
632 if(h->use_weight == 2){
633 int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
634 int weight1 = 64 - weight0;
635 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize,
636 height, 5, weight0, weight1, 0);
637 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
638 chroma_height, 5, weight0, weight1, 0);
639 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
640 chroma_height, 5, weight0, weight1, 0);
641 }else{
642 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height, h->luma_log2_weight_denom,
643 h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
644 h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
645 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
646 h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
647 h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
648 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
649 h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
650 h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
651 }
652 }else{
653 int list = list1 ? 1 : 0;
654 int refn = h->ref_cache[list][ scan8[n] ];
655 Picture *ref= &h->ref_list[list][refn];
656 mc_dir_part(h, ref, n, square, height, delta, list,
657 dest_y, dest_cb, dest_cr, x_offset, y_offset,
658 qpix_put, chroma_put, pixel_shift, chroma_idc);
659
660 luma_weight_op(dest_y, h->mb_linesize, height, h->luma_log2_weight_denom,
661 h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
662 if(h->use_weight_chroma){
663 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
664 h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
665 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
666 h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
667 }
668 }
669 }
670
671 static av_always_inline void
672 mc_part(H264Context *h, int n, int square, int height, int delta,
673 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
674 int x_offset, int y_offset,
675 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
676 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
677 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
678 int list0, int list1, int pixel_shift, int chroma_idc)
679 {
680 if((h->use_weight==2 && list0 && list1
681 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
682 || h->use_weight==1)
683 mc_part_weighted(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
684 x_offset, y_offset, qpix_put, chroma_put,
685 weight_op[0], weight_op[1], weight_avg[0],
686 weight_avg[1], list0, list1, pixel_shift, chroma_idc);
687 else
688 mc_part_std(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
689 x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
690 chroma_avg, list0, list1, pixel_shift, chroma_idc);
691 }
692
693 static av_always_inline void
694 prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc)
695 {
696 /* fetch pixels for estimated mv 4 macroblocks ahead
697 * optimized for 64byte cache lines */
698 MpegEncContext * const s = &h->s;
699 const int refn = h->ref_cache[list][scan8[0]];
700 if(refn >= 0){
701 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
702 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
703 uint8_t **src = h->ref_list[list][refn].f.data;
704 int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
705 s->dsp.prefetch(src[0]+off, s->linesize, 4);
706 if (chroma_idc == 3 /* yuv444 */) {
707 s->dsp.prefetch(src[1]+off, s->linesize, 4);
708 s->dsp.prefetch(src[2]+off, s->linesize, 4);
709 }else{
710 off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
711 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
712 }
713 }
714 }
715
716 static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
717 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
718 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
719 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
720 int pixel_shift, int chroma_idc)
721 {
722 MpegEncContext * const s = &h->s;
723 const int mb_xy= h->mb_xy;
724 const int mb_type = s->current_picture.f.mb_type[mb_xy];
725
726 assert(IS_INTER(mb_type));
727
728 if(HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
729 await_references(h);
730 prefetch_motion(h, 0, pixel_shift, chroma_idc);
731
732 if(IS_16X16(mb_type)){
733 mc_part(h, 0, 1, 16, 0, dest_y, dest_cb, dest_cr, 0, 0,
734 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
735 weight_op, weight_avg,
736 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
737 pixel_shift, chroma_idc);
738 }else if(IS_16X8(mb_type)){
739 mc_part(h, 0, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
740 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
741 weight_op, weight_avg,
742 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
743 pixel_shift, chroma_idc);
744 mc_part(h, 8, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
745 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
746 weight_op, weight_avg,
747 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
748 pixel_shift, chroma_idc);
749 }else if(IS_8X16(mb_type)){
750 mc_part(h, 0, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
751 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
752 &weight_op[1], &weight_avg[1],
753 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
754 pixel_shift, chroma_idc);
755 mc_part(h, 4, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
756 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
757 &weight_op[1], &weight_avg[1],
758 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
759 pixel_shift, chroma_idc);
760 }else{
761 int i;
762
763 assert(IS_8X8(mb_type));
764
765 for(i=0; i<4; i++){
766 const int sub_mb_type= h->sub_mb_type[i];
767 const int n= 4*i;
768 int x_offset= (i&1)<<2;
769 int y_offset= (i&2)<<1;
770
771 if(IS_SUB_8X8(sub_mb_type)){
772 mc_part(h, n, 1, 8, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
773 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
774 &weight_op[1], &weight_avg[1],
775 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
776 pixel_shift, chroma_idc);
777 }else if(IS_SUB_8X4(sub_mb_type)){
778 mc_part(h, n , 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
779 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
780 &weight_op[1], &weight_avg[1],
781 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
782 pixel_shift, chroma_idc);
783 mc_part(h, n+2, 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
784 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
785 &weight_op[1], &weight_avg[1],
786 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
787 pixel_shift, chroma_idc);
788 }else if(IS_SUB_4X8(sub_mb_type)){
789 mc_part(h, n , 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
790 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
791 &weight_op[2], &weight_avg[2],
792 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
793 pixel_shift, chroma_idc);
794 mc_part(h, n+1, 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
795 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
796 &weight_op[2], &weight_avg[2],
797 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
798 pixel_shift, chroma_idc);
799 }else{
800 int j;
801 assert(IS_SUB_4X4(sub_mb_type));
802 for(j=0; j<4; j++){
803 int sub_x_offset= x_offset + 2*(j&1);
804 int sub_y_offset= y_offset + (j&2);
805 mc_part(h, n+j, 1, 4, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
806 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
807 &weight_op[2], &weight_avg[2],
808 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
809 pixel_shift, chroma_idc);
810 }
811 }
812 }
813 }
814
815 prefetch_motion(h, 1, pixel_shift, chroma_idc);
816 }
817
818 static av_always_inline void
819 hl_motion_420(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
820 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
821 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
822 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
823 int pixel_shift)
824 {
825 hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
826 qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 1);
827 }
828
829 static av_always_inline void
830 hl_motion_422(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
831 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
832 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
833 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
834 int pixel_shift)
835 {
836 hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
837 qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 2);
838 }
839
840 static void free_tables(H264Context *h, int free_rbsp){
841 int i;
842 H264Context *hx;
843
844 av_freep(&h->intra4x4_pred_mode);
845 av_freep(&h->chroma_pred_mode_table);
846 av_freep(&h->cbp_table);
847 av_freep(&h->mvd_table[0]);
848 av_freep(&h->mvd_table[1]);
849 av_freep(&h->direct_table);
850 av_freep(&h->non_zero_count);
851 av_freep(&h->slice_table_base);
852 h->slice_table= NULL;
853 av_freep(&h->list_counts);
854
855 av_freep(&h->mb2b_xy);
856 av_freep(&h->mb2br_xy);
857
858 for(i = 0; i < MAX_THREADS; i++) {
859 hx = h->thread_context[i];
860 if(!hx) continue;
861 av_freep(&hx->top_borders[1]);
862 av_freep(&hx->top_borders[0]);
863 av_freep(&hx->s.obmc_scratchpad);
864 if (free_rbsp){
865 av_freep(&hx->rbsp_buffer[1]);
866 av_freep(&hx->rbsp_buffer[0]);
867 hx->rbsp_buffer_size[0] = 0;
868 hx->rbsp_buffer_size[1] = 0;
869 }
870 if (i) av_freep(&h->thread_context[i]);
871 }
872 }
873
874 static void init_dequant8_coeff_table(H264Context *h){
875 int i,j,q,x;
876 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
877
878 for(i=0; i<6; i++ ){
879 h->dequant8_coeff[i] = h->dequant8_buffer[i];
880 for(j=0; j<i; j++){
881 if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
882 h->dequant8_coeff[i] = h->dequant8_buffer[j];
883 break;
884 }
885 }
886 if(j<i)
887 continue;
888
889 for(q=0; q<max_qp+1; q++){
890 int shift = div6[q];
891 int idx = rem6[q];
892 for(x=0; x<64; x++)
893 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
894 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
895 h->pps.scaling_matrix8[i][x]) << shift;
896 }
897 }
898 }
899
900 static void init_dequant4_coeff_table(H264Context *h){
901 int i,j,q,x;
902 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
903 for(i=0; i<6; i++ ){
904 h->dequant4_coeff[i] = h->dequant4_buffer[i];
905 for(j=0; j<i; j++){
906 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
907 h->dequant4_coeff[i] = h->dequant4_buffer[j];
908 break;
909 }
910 }
911 if(j<i)
912 continue;
913
914 for(q=0; q<max_qp+1; q++){
915 int shift = div6[q] + 2;
916 int idx = rem6[q];
917 for(x=0; x<16; x++)
918 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
919 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
920 h->pps.scaling_matrix4[i][x]) << shift;
921 }
922 }
923 }
924
925 static void init_dequant_tables(H264Context *h){
926 int i,x;
927 init_dequant4_coeff_table(h);
928 if(h->pps.transform_8x8_mode)
929 init_dequant8_coeff_table(h);
930 if(h->sps.transform_bypass){
931 for(i=0; i<6; i++)
932 for(x=0; x<16; x++)
933 h->dequant4_coeff[i][0][x] = 1<<6;
934 if(h->pps.transform_8x8_mode)
935 for(i=0; i<6; i++)
936 for(x=0; x<64; x++)
937 h->dequant8_coeff[i][0][x] = 1<<6;
938 }
939 }
940
941
942 int ff_h264_alloc_tables(H264Context *h){
943 MpegEncContext * const s = &h->s;
944 const int big_mb_num= s->mb_stride * (s->mb_height+1);
945 const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
946 int x,y;
947
948 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8 * sizeof(uint8_t), fail)
949
950 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count , big_mb_num * 48 * sizeof(uint8_t), fail)
951 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
952 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
953
954 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
955 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
956 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
957 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
958 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
959
960 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
961 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
962
963 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy , big_mb_num * sizeof(uint32_t), fail);
964 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
965 for(y=0; y<s->mb_height; y++){
966 for(x=0; x<s->mb_width; x++){
967 const int mb_xy= x + y*s->mb_stride;
968 const int b_xy = 4*x + 4*y*h->b_stride;
969
970 h->mb2b_xy [mb_xy]= b_xy;
971 h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
972 }
973 }
974
975 s->obmc_scratchpad = NULL;
976
977 if(!h->dequant4_coeff[0])
978 init_dequant_tables(h);
979
980 return 0;
981 fail:
982 free_tables(h, 1);
983 return -1;
984 }
985
986 /**
987 * Mimic alloc_tables(), but for every context thread.
988 */
989 static void clone_tables(H264Context *dst, H264Context *src, int i){
990 MpegEncContext * const s = &src->s;
991 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
992 dst->non_zero_count = src->non_zero_count;
993 dst->slice_table = src->slice_table;
994 dst->cbp_table = src->cbp_table;
995 dst->mb2b_xy = src->mb2b_xy;
996 dst->mb2br_xy = src->mb2br_xy;
997 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
998 dst->mvd_table[0] = src->mvd_table[0] + i*8*2*s->mb_stride;
999 dst->mvd_table[1] = src->mvd_table[1] + i*8*2*s->mb_stride;
1000 dst->direct_table = src->direct_table;
1001 dst->list_counts = src->list_counts;
1002
1003 dst->s.obmc_scratchpad = NULL;
1004 ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma, src->sps.chroma_format_idc);
1005 }
1006
1007 /**
1008 * Init context
1009 * Allocate buffers which are not shared amongst multiple threads.
1010 */
1011 static int context_init(H264Context *h){
1012 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1013 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1014
1015 h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
1016 h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
1017
1018 return 0;
1019 fail:
1020 return -1; // free_tables will clean up for us
1021 }
1022
1023 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
1024
1025 static av_cold void common_init(H264Context *h){
1026 MpegEncContext * const s = &h->s;
1027
1028 s->width = s->avctx->width;
1029 s->height = s->avctx->height;
1030 s->codec_id= s->avctx->codec->id;
1031
1032 ff_h264dsp_init(&h->h264dsp, 8, 1);
1033 ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
1034
1035 h->dequant_coeff_pps= -1;
1036 s->unrestricted_mv=1;
1037
1038 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
1039
1040 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
1041 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
1042 }
1043
1044 int ff_h264_decode_extradata(H264Context *h)
1045 {
1046 AVCodecContext *avctx = h->s.avctx;
1047
1048 if(avctx->extradata[0] == 1){
1049 int i, cnt, nalsize;
1050 unsigned char *p = avctx->extradata;
1051
1052 h->is_avc = 1;
1053
1054 if(avctx->extradata_size < 7) {
1055 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1056 return -1;
1057 }
1058 /* sps and pps in the avcC always have length coded with 2 bytes,
1059 so put a fake nal_length_size = 2 while parsing them */
1060 h->nal_length_size = 2;
1061 // Decode sps from avcC
1062 cnt = *(p+5) & 0x1f; // Number of sps
1063 p += 6;
1064 for (i = 0; i < cnt; i++) {
1065 nalsize = AV_RB16(p) + 2;
1066 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1067 return -1;
1068 if(decode_nal_units(h, p, nalsize) < 0) {
1069 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1070 return -1;
1071 }
1072 p += nalsize;
1073 }
1074 // Decode pps from avcC
1075 cnt = *(p++); // Number of pps
1076 for (i = 0; i < cnt; i++) {
1077 nalsize = AV_RB16(p) + 2;
1078 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1079 return -1;
1080 if (decode_nal_units(h, p, nalsize) < 0) {
1081 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1082 return -1;
1083 }
1084 p += nalsize;
1085 }
1086 // Now store right nal length size, that will be use to parse all other nals
1087 h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
1088 } else {
1089 h->is_avc = 0;
1090 if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
1091 return -1;
1092 }
1093 return 0;
1094 }
1095
1096 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1097 H264Context *h= avctx->priv_data;
1098 MpegEncContext * const s = &h->s;
1099 int i;
1100
1101 MPV_decode_defaults(s);
1102
1103 s->avctx = avctx;
1104 common_init(h);
1105
1106 s->out_format = FMT_H264;
1107 s->workaround_bugs= avctx->workaround_bugs;
1108
1109 // set defaults
1110 // s->decode_mb= ff_h263_decode_mb;
1111 s->quarter_sample = 1;
1112 if(!avctx->has_b_frames)
1113 s->low_delay= 1;
1114
1115 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1116
1117 ff_h264_decode_init_vlc();
1118
1119 h->pixel_shift = 0;
1120 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1121
1122 h->thread_context[0] = h;
1123 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1124 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1125 h->last_pocs[i] = INT_MIN;
1126 h->prev_poc_msb= 1<<16;
1127 h->x264_build = -1;
1128 ff_h264_reset_sei(h);
1129 if(avctx->codec_id == CODEC_ID_H264){
1130 if(avctx->ticks_per_frame == 1){
1131 s->avctx->time_base.den *=2;
1132 }
1133 avctx->ticks_per_frame = 2;
1134 }
1135
1136 if(avctx->extradata_size > 0 && avctx->extradata &&
1137 ff_h264_decode_extradata(h))
1138 return -1;
1139
1140 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1141 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1142 s->low_delay = 0;
1143 }
1144
1145 return 0;
1146 }
1147
1148 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1149 static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1150 {
1151 int i;
1152
1153 for (i=0; i<count; i++){
1154 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1155 IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1156 !from[i]));
1157 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1158 }
1159 }
1160
1161 static void copy_parameter_set(void **to, void **from, int count, int size)
1162 {
1163 int i;
1164
1165 for (i=0; i<count; i++){
1166 if (to[i] && !from[i]) av_freep(&to[i]);
1167 else if (from[i] && !to[i]) to[i] = av_malloc(size);
1168
1169 if (from[i]) memcpy(to[i], from[i], size);
1170 }
1171 }
1172
1173 static int decode_init_thread_copy(AVCodecContext *avctx){
1174 H264Context *h= avctx->priv_data;
1175
1176 if (!avctx->internal->is_copy)
1177 return 0;
1178 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1179 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1180
1181 return 0;
1182 }
1183
1184 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1185 static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1186 H264Context *h= dst->priv_data, *h1= src->priv_data;
1187 MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1188 int inited = s->context_initialized, err;
1189 int i;
1190
1191 if(dst == src || !s1->context_initialized) return 0;
1192
1193 err = ff_mpeg_update_thread_context(dst, src);
1194 if(err) return err;
1195
1196 //FIXME handle width/height changing
1197 if(!inited){
1198 for(i = 0; i < MAX_SPS_COUNT; i++)
1199 av_freep(h->sps_buffers + i);
1200
1201 for(i = 0; i < MAX_PPS_COUNT; i++)
1202 av_freep(h->pps_buffers + i);
1203
1204 memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1205 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1206 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1207 if (ff_h264_alloc_tables(h) < 0) {
1208 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1209 return AVERROR(ENOMEM);
1210 }
1211 context_init(h);
1212
1213 for(i=0; i<2; i++){
1214 h->rbsp_buffer[i] = NULL;
1215 h->rbsp_buffer_size[i] = 0;
1216 }
1217
1218 h->thread_context[0] = h;
1219
1220 // frame_start may not be called for the next thread (if it's decoding a bottom field)
1221 // so this has to be allocated here
1222 h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1223
1224 s->dsp.clear_blocks(h->mb);
1225 s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
1226 }
1227
1228 //extradata/NAL handling
1229 h->is_avc = h1->is_avc;
1230
1231 //SPS/PPS
1232 copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1233 h->sps = h1->sps;
1234 copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1235 h->pps = h1->pps;
1236
1237 //Dequantization matrices
1238 //FIXME these are big - can they be only copied when PPS changes?
1239 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1240
1241 for(i=0; i<6; i++)
1242 h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1243
1244 for(i=0; i<6; i++)
1245 h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1246
1247 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1248
1249 //POC timing
1250 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1251
1252 //reference lists
1253 copy_fields(h, h1, ref_count, list_count);
1254 copy_fields(h, h1, ref_list, intra_gb);
1255 copy_fields(h, h1, short_ref, cabac_init_idc);
1256
1257 copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1258 copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1259 copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1260
1261 h->last_slice_type = h1->last_slice_type;
1262
1263 if(!s->current_picture_ptr) return 0;
1264
1265 if(!s->dropable) {
1266 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1267 h->prev_poc_msb = h->poc_msb;
1268 h->prev_poc_lsb = h->poc_lsb;
1269 }
1270 h->prev_frame_num_offset= h->frame_num_offset;
1271 h->prev_frame_num = h->frame_num;
1272 h->outputed_poc = h->next_outputed_poc;
1273
1274 return err;
1275 }
1276
1277 int ff_h264_frame_start(H264Context *h){
1278 MpegEncContext * const s = &h->s;
1279 int i;
1280 const int pixel_shift = h->pixel_shift;
1281 int thread_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1;
1282
1283 if(MPV_frame_start(s, s->avctx) < 0)
1284 return -1;
1285 ff_er_frame_start(s);
1286 /*
1287 * MPV_frame_start uses pict_type to derive key_frame.
1288 * This is incorrect for H.264; IDR markings must be used.
1289 * Zero here; IDR markings per slice in frame or fields are ORed in later.
1290 * See decode_nal_units().
1291 */
1292 s->current_picture_ptr->f.key_frame = 0;
1293 s->current_picture_ptr->mmco_reset= 0;
1294
1295 assert(s->linesize && s->uvlinesize);
1296
1297 for(i=0; i<16; i++){
1298 h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
1299 h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
1300 }
1301 for(i=0; i<16; i++){
1302 h->block_offset[16+i]=
1303 h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1304 h->block_offset[48+16+i]=
1305 h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1306 }
1307
1308 /* can't be in alloc_tables because linesize isn't known there.
1309 * FIXME: redo bipred weight to not require extra buffer? */
1310 for(i = 0; i < thread_count; i++)
1311 if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1312 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1313
1314 /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
1315 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
1316
1317 // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.f.reference /*|| h->contains_intra*/ || 1;
1318
1319 // We mark the current picture as non-reference after allocating it, so
1320 // that if we break out due to an error it can be released automatically
1321 // in the next MPV_frame_start().
1322 // SVQ3 as well as most other codecs have only last/next/current and thus
1323 // get released even with set reference, besides SVQ3 and others do not
1324 // mark frames as reference later "naturally".
1325 if(s->codec_id != CODEC_ID_SVQ3)
1326 s->current_picture_ptr->f.reference = 0;
1327
1328 s->current_picture_ptr->field_poc[0]=
1329 s->current_picture_ptr->field_poc[1]= INT_MAX;
1330
1331 h->next_output_pic = NULL;
1332
1333 assert(s->current_picture_ptr->long_ref==0);
1334
1335 return 0;
1336 }
1337
1338 /**
1339 * Run setup operations that must be run after slice header decoding.
1340 * This includes finding the next displayed frame.
1341 *
1342 * @param h h264 master context
1343 * @param setup_finished enough NALs have been read that we can call
1344 * ff_thread_finish_setup()
1345 */
1346 static void decode_postinit(H264Context *h, int setup_finished){
1347 MpegEncContext * const s = &h->s;
1348 Picture *out = s->current_picture_ptr;
1349 Picture *cur = s->current_picture_ptr;
1350 int i, pics, out_of_order, out_idx;
1351 int invalid = 0, cnt = 0;
1352
1353 s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1354 s->current_picture_ptr->f.pict_type = s->pict_type;
1355
1356 if (h->next_output_pic) return;
1357
1358 if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1359 //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1360 //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1361 //to find this yet, so we assume the worst for now.
1362 //if (setup_finished)
1363 // ff_thread_finish_setup(s->avctx);
1364 return;
1365 }
1366
1367 cur->f.interlaced_frame = 0;
1368 cur->f.repeat_pict = 0;
1369
1370 /* Signal interlacing information externally. */
1371 /* Prioritize picture timing SEI information over used decoding process if it exists. */
1372
1373 if(h->sps.pic_struct_present_flag){
1374 switch (h->sei_pic_struct)
1375 {
1376 case SEI_PIC_STRUCT_FRAME:
1377 break;
1378 case SEI_PIC_STRUCT_TOP_FIELD:
1379 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1380 cur->f.interlaced_frame = 1;
1381 break;
1382 case SEI_PIC_STRUCT_TOP_BOTTOM:
1383 case SEI_PIC_STRUCT_BOTTOM_TOP:
1384 if (FIELD_OR_MBAFF_PICTURE)
1385 cur->f.interlaced_frame = 1;
1386 else
1387 // try to flag soft telecine progressive
1388 cur->f.interlaced_frame = h->prev_interlaced_frame;
1389 break;
1390 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1391 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1392 // Signal the possibility of telecined film externally (pic_struct 5,6)
1393 // From these hints, let the applications decide if they apply deinterlacing.
1394 cur->f.repeat_pict = 1;
1395 break;
1396 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1397 // Force progressive here, as doubling interlaced frame is a bad idea.
1398 cur->f.repeat_pict = 2;
1399 break;
1400 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1401 cur->f.repeat_pict = 4;
1402 break;
1403 }
1404
1405 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1406 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1407 }else{
1408 /* Derive interlacing flag from used decoding process. */
1409 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1410 }
1411 h->prev_interlaced_frame = cur->f.interlaced_frame;
1412
1413 if (cur->field_poc[0] != cur->field_poc[1]){
1414 /* Derive top_field_first from field pocs. */
1415 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1416 }else{
1417 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1418 /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1419 if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1420 || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1421 cur->f.top_field_first = 1;
1422 else
1423 cur->f.top_field_first = 0;
1424 }else{
1425 /* Most likely progressive */
1426 cur->f.top_field_first = 0;
1427 }
1428 }
1429
1430 //FIXME do something with unavailable reference frames
1431
1432 /* Sort B-frames into display order */
1433
1434 if(h->sps.bitstream_restriction_flag
1435 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1436 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1437 s->low_delay = 0;
1438 }
1439
1440 if( s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1441 && !h->sps.bitstream_restriction_flag){
1442 s->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
1443 s->low_delay= 0;
1444 }
1445
1446 pics = 0;
1447 while(h->delayed_pic[pics]) pics++;
1448
1449 assert(pics <= MAX_DELAYED_PIC_COUNT);
1450
1451 h->delayed_pic[pics++] = cur;
1452 if (cur->f.reference == 0)
1453 cur->f.reference = DELAYED_PIC_REF;
1454
1455 /* Frame reordering. This code takes pictures from coding order and sorts
1456 * them by their incremental POC value into display order. It supports POC
1457 * gaps, MMCO reset codes and random resets.
1458 * A "display group" can start either with a IDR frame (f.key_frame = 1),
1459 * and/or can be closed down with a MMCO reset code. In sequences where
1460 * there is no delay, we can't detect that (since the frame was already
1461 * output to the user), so we also set h->mmco_reset to detect the MMCO
1462 * reset code.
1463 * FIXME: if we detect insufficient delays (as per s->avctx->has_b_frames),
1464 * we increase the delay between input and output. All frames affected by
1465 * the lag (e.g. those that should have been output before another frame
1466 * that we already returned to the user) will be dropped. This is a bug
1467 * that we will fix later. */
1468 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
1469 cnt += out->poc < h->last_pocs[i];
1470 invalid += out->poc == INT_MIN;
1471 }
1472 if (!h->mmco_reset && !cur->f.key_frame && cnt + invalid == MAX_DELAYED_PIC_COUNT && cnt > 0) {
1473 h->mmco_reset = 2;
1474 if (pics > 1)
1475 h->delayed_pic[pics - 2]->mmco_reset = 2;
1476 }
1477 if (h->mmco_reset || cur->f.key_frame) {
1478 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1479 h->last_pocs[i] = INT_MIN;
1480 cnt = 0;
1481 invalid = MAX_DELAYED_PIC_COUNT;
1482 }
1483 out = h->delayed_pic[0];
1484 out_idx = 0;
1485 for (i = 1; i < MAX_DELAYED_PIC_COUNT && h->delayed_pic[i] &&
1486 !h->delayed_pic[i-1]->mmco_reset && !h->delayed_pic[i]->f.key_frame; i++)
1487 {
1488 if(h->delayed_pic[i]->poc < out->poc){
1489 out = h->delayed_pic[i];
1490 out_idx = i;
1491 }
1492 }
1493 if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->mmco_reset))
1494 h->next_outputed_poc = INT_MIN;
1495 out_of_order = !out->f.key_frame && !h->mmco_reset && (out->poc < h->next_outputed_poc);
1496
1497 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
1498 { }
1499 else if (out_of_order && pics-1 == s->avctx->has_b_frames &&
1500 s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
1501 if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
1502 s->avctx->has_b_frames = FFMAX(s->avctx->has_b_frames, cnt);
1503 }
1504 s->low_delay = 0;
1505 } else if (s->low_delay &&
1506 ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2) ||
1507 cur->f.pict_type == AV_PICTURE_TYPE_B)) {
1508 s->low_delay = 0;
1509 s->avctx->has_b_frames++;
1510 }
1511
1512 if(pics > s->avctx->has_b_frames){
1513 out->f.reference &= ~DELAYED_PIC_REF;
1514 out->owner2 = s; // for frame threading, the owner must be the second field's thread
1515 // or else the first thread can release the picture and reuse it unsafely
1516 for(i=out_idx; h->delayed_pic[i]; i++)
1517 h->delayed_pic[i] = h->delayed_pic[i+1];
1518 }
1519 memmove(h->last_pocs, &h->last_pocs[1], sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
1520 h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = cur->poc;
1521 if(!out_of_order && pics > s->avctx->has_b_frames){
1522 h->next_output_pic = out;
1523 if (out->mmco_reset) {
1524 if (out_idx > 0) {
1525 h->next_outputed_poc = out->poc;
1526 h->delayed_pic[out_idx - 1]->mmco_reset = out->mmco_reset;
1527 } else {
1528 h->next_outputed_poc = INT_MIN;
1529 }
1530 } else {
1531 h->next_outputed_poc = out->poc;
1532 }
1533 h->mmco_reset = 0;
1534 }else{
1535 av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1536 }
1537
1538 if (setup_finished)
1539 ff_thread_finish_setup(s->avctx);
1540 }
1541
1542 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1543 uint8_t *src_cb, uint8_t *src_cr,
1544 int linesize, int uvlinesize, int simple)
1545 {
1546 MpegEncContext * const s = &h->s;
1547 uint8_t *top_border;
1548 int top_idx = 1;
1549 const int pixel_shift = h->pixel_shift;
1550 int chroma444 = CHROMA444;
1551 int chroma422 = CHROMA422;
1552
1553 src_y -= linesize;
1554 src_cb -= uvlinesize;
1555 src_cr -= uvlinesize;
1556
1557 if(!simple && FRAME_MBAFF){
1558 if(s->mb_y&1){
1559 if(!MB_MBAFF){
1560 top_border = h->top_borders[0][s->mb_x];
1561 AV_COPY128(top_border, src_y + 15*linesize);
1562 if (pixel_shift)
1563 AV_COPY128(top_border+16, src_y+15*linesize+16);
1564 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1565 if(chroma444){
1566 if (pixel_shift){
1567 AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1568 AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1569 AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1570 AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1571 } else {
1572 AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1573 AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1574 }
1575 } else if(chroma422) {
1576 if (pixel_shift) {
1577 AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1578 AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
1579 } else {
1580 AV_COPY64(top_border+16, src_cb + 15*uvlinesize);
1581 AV_COPY64(top_border+24, src_cr + 15*uvlinesize);
1582 }
1583 } else {
1584 if (pixel_shift) {
1585 AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1586 AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1587 } else {
1588 AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1589 AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1590 }
1591 }
1592 }
1593 }
1594 }else if(MB_MBAFF){
1595 top_idx = 0;
1596 }else
1597 return;
1598 }
1599
1600 top_border = h->top_borders[top_idx][s->mb_x];
1601 // There are two lines saved, the line above the the top macroblock of a pair,
1602 // and the line above the bottom macroblock
1603 AV_COPY128(top_border, src_y + 16*linesize);
1604 if (pixel_shift)
1605 AV_COPY128(top_border+16, src_y+16*linesize+16);
1606
1607 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1608 if(chroma444){
1609 if (pixel_shift){
1610 AV_COPY128(top_border+32, src_cb + 16*linesize);
1611 AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1612 AV_COPY128(top_border+64, src_cr + 16*linesize);
1613 AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1614 } else {
1615 AV_COPY128(top_border+16, src_cb + 16*linesize);
1616 AV_COPY128(top_border+32, src_cr + 16*linesize);
1617 }
1618 } else if(chroma422) {
1619 if (pixel_shift) {
1620 AV_COPY128(top_border+32, src_cb+16*uvlinesize);
1621 AV_COPY128(top_border+48, src_cr+16*uvlinesize);
1622 } else {
1623 AV_COPY64(top_border+16, src_cb+16*uvlinesize);
1624 AV_COPY64(top_border+24, src_cr+16*uvlinesize);
1625 }
1626 } else {
1627 if (pixel_shift) {
1628 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1629 AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1630 } else {
1631 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1632 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1633 }
1634 }
1635 }
1636 }
1637
1638 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1639 uint8_t *src_cb, uint8_t *src_cr,
1640 int linesize, int uvlinesize,
1641 int xchg, int chroma444,
1642 int simple, int pixel_shift){
1643 MpegEncContext * const s = &h->s;
1644 int deblock_topleft;
1645 int deblock_top;
1646 int top_idx = 1;
1647 uint8_t *top_border_m1;
1648 uint8_t *top_border;
1649
1650 if(!simple && FRAME_MBAFF){
1651 if(s->mb_y&1){
1652 if(!MB_MBAFF)
1653 return;
1654 }else{
1655 top_idx = MB_MBAFF ? 0 : 1;
1656 }
1657 }
1658
1659 if(h->deblocking_filter == 2) {
1660 deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1661 deblock_top = h->top_type;
1662 } else {
1663 deblock_topleft = (s->mb_x > 0);
1664 deblock_top = (s->mb_y > !!MB_FIELD);
1665 }
1666
1667 src_y -= linesize + 1 + pixel_shift;
1668 src_cb -= uvlinesize + 1 + pixel_shift;
1669 src_cr -= uvlinesize + 1 + pixel_shift;
1670
1671 top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1672 top_border = h->top_borders[top_idx][s->mb_x];
1673
1674 #define XCHG(a,b,xchg)\
1675 if (pixel_shift) {\
1676 if (xchg) {\
1677 AV_SWAP64(b+0,a+0);\
1678 AV_SWAP64(b+8,a+8);\
1679 } else {\
1680 AV_COPY128(b,a); \
1681 }\
1682 } else \
1683 if (xchg) AV_SWAP64(b,a);\
1684 else AV_COPY64(b,a);
1685
1686 if(deblock_top){
1687 if(deblock_topleft){
1688 XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
1689 }
1690 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1691 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1692 if(s->mb_x+1 < s->mb_width){
1693 XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
1694 }
1695 }
1696 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1697 if(chroma444){
1698 if(deblock_topleft){
1699 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1700 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1701 }
1702 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1703 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1704 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1705 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1706 if(s->mb_x+1 < s->mb_width){
1707 XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1708 XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1709 }
1710 } else {
1711 if(deblock_top){
1712 if(deblock_topleft){
1713 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1714 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1715 }
1716 XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1717 XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
1718 }
1719 }
1720 }
1721 }
1722
1723 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1724 if (high_bit_depth) {
1725 return AV_RN32A(((int32_t*)mb) + index);
1726 } else
1727 return AV_RN16A(mb + index);
1728 }
1729
1730 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1731 if (high_bit_depth) {
1732 AV_WN32A(((int32_t*)mb) + index, value);
1733 } else
1734 AV_WN16A(mb + index, value);
1735 }
1736
1737 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1738 int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1739 {
1740 MpegEncContext * const s = &h->s;
1741 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1742 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1743 int i;
1744 int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1745 block_offset += 16*p;
1746 if(IS_INTRA4x4(mb_type)){
1747 if(simple || !s->encoding){
1748 if(IS_8x8DCT(mb_type)){
1749 if(transform_bypass){
1750 idct_dc_add =
1751 idct_add = s->dsp.add_pixels8;
1752 }else{
1753 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1754 idct_add = h->h264dsp.h264_idct8_add;
1755 }
1756 for(i=0; i<16; i+=4){
1757 uint8_t * const ptr= dest_y + block_offset[i];
1758 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1759 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1760 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1761 }else{
1762 const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1763 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1764 (h->topright_samples_available<<i)&0x4000, linesize);
1765 if(nnz){
1766 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1767 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1768 else
1769 idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1770 }
1771 }
1772 }
1773 }else{
1774 if(transform_bypass){
1775 idct_dc_add =
1776 idct_add = s->dsp.add_pixels4;
1777 }else{
1778 idct_dc_add = h->h264dsp.h264_idct_dc_add;
1779 idct_add = h->h264dsp.h264_idct_add;
1780 }
1781 for(i=0; i<16; i++){
1782 uint8_t * const ptr= dest_y + block_offset[i];
1783 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1784
1785 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1786 h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1787 }else{
1788 uint8_t *topright;
1789 int nnz, tr;
1790 uint64_t tr_high;
1791 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1792 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1793 assert(s->mb_y || linesize <= block_offset[i]);
1794 if(!topright_avail){
1795 if (pixel_shift) {
1796 tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1797 topright= (uint8_t*) &tr_high;
1798 } else {
1799 tr= ptr[3 - linesize]*0x01010101u;
1800 topright= (uint8_t*) &tr;
1801 }
1802 }else
1803 topright= ptr + (4 << pixel_shift) - linesize;
1804 }else
1805 topright= NULL;
1806
1807 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1808 nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1809 if(nnz){
1810 if(is_h264){
1811 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1812 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1813 else
1814 idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1815 }else
1816 ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1817 }
1818 }
1819 }
1820 }
1821 }
1822 }else{
1823 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1824 if(is_h264){
1825 if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
1826 if(!transform_bypass)
1827 h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1828 else{
1829 static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1830 8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1831 for(i = 0; i < 16; i++)
1832 dctcoef_set(h->mb+(p*256 << pixel_shift), pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
1833 }
1834 }
1835 }else
1836 ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1837 }
1838 }
1839
1840 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1841 int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1842 {
1843 MpegEncContext * const s = &h->s;
1844 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1845 int i;
1846 block_offset += 16*p;
1847 if(!IS_INTRA4x4(mb_type)){
1848 if(is_h264){
1849 if(IS_INTRA16x16(mb_type)){
1850 if(transform_bypass){
1851 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1852 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1853 }else{
1854 for(i=0; i<16; i++){
1855 if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1856 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1857 }
1858 }
1859 }else{
1860 h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1861 }
1862 }else if(h->cbp&15){
1863 if(transform_bypass){
1864 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1865 idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1866 for(i=0; i<16; i+=di){
1867 if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1868 idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1869 }
1870 }
1871 }else{
1872 if(IS_8x8DCT(mb_type)){
1873 h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1874 }else{
1875 h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1876 }
1877 }
1878 }
1879 }else{
1880 for(i=0; i<16; i++){
1881 if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1882 uint8_t * const ptr= dest_y + block_offset[i];
1883 ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1884 }
1885 }
1886 }
1887 }
1888 }
1889
1890 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift)
1891 {
1892 MpegEncContext * const s = &h->s;
1893 const int mb_x= s->mb_x;
1894 const int mb_y= s->mb_y;
1895 const int mb_xy= h->mb_xy;
1896 const int mb_type = s->current_picture.f.mb_type[mb_xy];
1897 uint8_t *dest_y, *dest_cb, *dest_cr;
1898 int linesize, uvlinesize /*dct_offset*/;
1899 int i, j;
1900 int *block_offset = &h->block_offset[0];
1901 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1902 /* is_h264 should always be true if SVQ3 is disabled. */
1903 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1904 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1905 const int block_h = 16 >> s->chroma_y_shift;
1906 const int chroma422 = CHROMA422;
1907
1908 dest_y = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
1909 dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1910 dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1911
1912 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1913 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
1914
1915 h->list_counts[mb_xy]= h->list_count;
1916
1917 if (!simple && MB_FIELD) {
1918 linesize = h->mb_linesize = s->linesize * 2;
1919 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1920 block_offset = &h->block_offset[48];
1921 if(mb_y&1){ //FIXME move out of this function?
1922 dest_y -= s->linesize*15;
1923 dest_cb-= s->uvlinesize * (block_h - 1);
1924 dest_cr-= s->uvlinesize * (block_h - 1);
1925 }
1926 if(FRAME_MBAFF) {
1927 int list;
1928 for(list=0; list<h->list_count; list++){
1929 if(!USES_LIST(mb_type, list))
1930 continue;
1931 if(IS_16X16(mb_type)){
1932 int8_t *ref = &h->ref_cache[list][scan8[0]];
1933 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1934 }else{
1935 for(i=0; i<16; i+=4){
1936 int ref = h->ref_cache[list][scan8[i]];
1937 if(ref >= 0)
1938 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1939 }
1940 }
1941 }
1942 }
1943 } else {
1944 linesize = h->mb_linesize = s->linesize;
1945 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1946 // dct_offset = s->linesize * 16;
1947 }
1948
1949 if (!simple && IS_INTRA_PCM(mb_type)) {
1950 if (pixel_shift) {
1951 const int bit_depth = h->sps.bit_depth_luma;
1952 int j;
1953 GetBitContext gb;
1954 init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1955
1956 for (i = 0; i < 16; i++) {
1957 uint16_t *tmp_y = (uint16_t*)(dest_y + i*linesize);
1958 for (j = 0; j < 16; j++)
1959 tmp_y[j] = get_bits(&gb, bit_depth);
1960 }
1961 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1962 if (!h->sps.chroma_format_idc) {
1963 for (i = 0; i < block_h; i++) {
1964 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1965 for (j = 0; j < 8; j++) {
1966 tmp_cb[j] = 1 << (bit_depth - 1);
1967 }
1968 }
1969 for (i = 0; i < block_h; i++) {
1970 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1971 for (j = 0; j < 8; j++) {
1972 tmp_cr[j] = 1 << (bit_depth - 1);
1973 }
1974 }
1975 } else {
1976 for (i = 0; i < block_h; i++) {
1977 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1978 for (j = 0; j < 8; j++)
1979 tmp_cb[j] = get_bits(&gb, bit_depth);
1980 }
1981 for (i = 0; i < block_h; i++) {
1982 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1983 for (j = 0; j < 8; j++)
1984 tmp_cr[j] = get_bits(&gb, bit_depth);
1985 }
1986 }
1987 }
1988 } else {
1989 for (i=0; i<16; i++) {
1990 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
1991 }
1992 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1993 if (!h->sps.chroma_format_idc) {
1994 for (i = 0; i < block_h; i++) {
1995 memset(dest_cb + i*uvlinesize, 128, 8);
1996 memset(dest_cr + i*uvlinesize, 128, 8);
1997 }
1998 } else {
1999 for (i = 0; i < block_h; i++) {
2000 memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4, 8);
2001 memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4, 8);
2002 }
2003 }
2004 }
2005 }
2006 } else {
2007 if(IS_INTRA(mb_type)){
2008 if(h->deblocking_filter)
2009 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
2010
2011 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2012 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2013 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2014 }
2015
2016 hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
2017
2018 if(h->deblocking_filter)
2019 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
2020 }else if(is_h264){
2021 if (chroma422) {
2022 hl_motion_422(h, dest_y, dest_cb, dest_cr,
2023 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2024 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2025 h->h264dsp.weight_h264_pixels_tab,
2026 h->h264dsp.biweight_h264_pixels_tab,
2027 pixel_shift);
2028 } else {
2029 hl_motion_420(h, dest_y, dest_cb, dest_cr,
2030 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2031 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2032 h->h264dsp.weight_h264_pixels_tab,
2033 h->h264dsp.biweight_h264_pixels_tab,
2034 pixel_shift);
2035 }
2036 }
2037
2038 hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
2039
2040 if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2041 uint8_t *dest[2] = {dest_cb, dest_cr};
2042 if(transform_bypass){
2043 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2044 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
2045 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
2046 }else{
2047 idct_add = s->dsp.add_pixels4;
2048 for(j=1; j<3; j++){
2049 for(i=j*16; i<j*16+4; i++){
2050 if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2051 idct_add (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
2052 }
2053 if (chroma422) {
2054 for(i=j*16+4; i<j*16+8; i++){
2055 if(h->non_zero_count_cache[ scan8[i+4] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2056 idct_add (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
2057 }
2058 }
2059 }
2060 }
2061 }else{
2062 if(is_h264){
2063 int qp[2];
2064 if (chroma422) {
2065 qp[0] = h->chroma_qp[0] + 3;
2066 qp[1] = h->chroma_qp[1] + 3;
2067 } else {
2068 qp[0] = h->chroma_qp[0];
2069 qp[1] = h->chroma_qp[1];
2070 }
2071 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
2072 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*1 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][qp[0]][0]);
2073 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
2074 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*2 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][qp[1]][0]);
2075 h->h264dsp.h264_idct_add8(dest, block_offset,
2076 h->mb, uvlinesize,
2077 h->non_zero_count_cache);
2078 }else{
2079 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*1, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
2080 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*2, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
2081 for(j=1; j<3; j++){
2082 for(i=j*16; i<j*16+4; i++){
2083 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2084 uint8_t * const ptr= dest[j-1] + block_offset[i];
2085 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
2086 }
2087 }
2088 }
2089 }
2090 }
2091 }
2092 }
2093 if(h->cbp || IS_INTRA(mb_type))
2094 {
2095 s->dsp.clear_blocks(h->mb);
2096 s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2097 }
2098 }
2099
2100 static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
2101 MpegEncContext * const s = &h->s;
2102 const int mb_x= s->mb_x;
2103 const int mb_y= s->mb_y;
2104 const int mb_xy= h->mb_xy;
2105 const int mb_type = s->current_picture.f.mb_type[mb_xy];
2106 uint8_t *dest[3];
2107 int linesize;
2108 int i, j, p;
2109 int *block_offset = &h->block_offset[0];
2110 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2111 const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
2112
2113 for (p = 0; p < plane_count; p++)
2114 {
2115 dest[p] = s->current_picture.f.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
2116 s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
2117 }
2118
2119 h->list_counts[mb_xy]= h->list_count;
2120
2121 if (!simple && MB_FIELD) {
2122 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
2123 block_offset = &h->block_offset[48];
2124 if(mb_y&1) //FIXME move out of this function?
2125 for (p = 0; p < 3; p++)
2126 dest[p] -= s->linesize*15;
2127 if(FRAME_MBAFF) {
2128 int list;
2129 for(list=0; list<h->list_count; list++){
2130 if(!USES_LIST(mb_type, list))
2131 continue;
2132 if(IS_16X16(mb_type)){
2133 int8_t *ref = &h->ref_cache[list][scan8[0]];
2134 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2135 }else{
2136 for(i=0; i<16; i+=4){
2137 int ref = h->ref_cache[list][scan8[i]];
2138 if(ref >= 0)
2139 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2140 }
2141 }
2142 }
2143 }
2144 } else {
2145 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2146 }
2147
2148 if (!simple && IS_INTRA_PCM(mb_type)) {
2149 if (pixel_shift) {
2150 const int bit_depth = h->sps.bit_depth_luma;
2151 GetBitContext gb;
2152 init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2153
2154 for (p = 0; p < plane_count; p++) {
2155 for (i = 0; i < 16; i++) {
2156 uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2157 for (j = 0; j < 16; j++)
2158 tmp[j] = get_bits(&gb, bit_depth);
2159 }
2160 }
2161 } else {
2162 for (p = 0; p < plane_count; p++) {
2163 for (i = 0; i < 16; i++) {
2164 memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2165 }
2166 }
2167 }
2168 } else {
2169 if(IS_INTRA(mb_type)){
2170 if(h->deblocking_filter)
2171 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2172
2173 for (p = 0; p < plane_count; p++)
2174 hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2175
2176 if(h->deblocking_filter)
2177 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2178 }else{
2179 hl_motion(h, dest[0], dest[1], dest[2],
2180 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2181 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2182 h->h264dsp.weight_h264_pixels_tab,
2183 h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 3);
2184 }
2185
2186 for (p = 0; p < plane_count; p++)
2187 hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2188 }
2189 if(h->cbp || IS_INTRA(mb_type))
2190 {
2191 s->dsp.clear_blocks(h->mb);
2192 s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2193 }
2194 }
2195
2196 /**
2197 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2198 */
2199 #define hl_decode_mb_simple(sh, bits) \
2200 static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2201 hl_decode_mb_internal(h, 1, sh); \
2202 }
2203 hl_decode_mb_simple(0, 8);
2204 hl_decode_mb_simple(1, 16);
2205
2206 /**
2207 * Process a macroblock; this handles edge cases, such as interlacing.
2208 */
2209 static void av_noinline hl_decode_mb_complex(H264Context *h){
2210 hl_decode_mb_internal(h, 0, h->pixel_shift);
2211 }
2212
2213 static void av_noinline hl_decode_mb_444_complex(H264Context *h){
2214 hl_decode_mb_444_internal(h, 0, h->pixel_shift);
2215 }
2216
2217 static void av_noinline hl_decode_mb_444_simple(H264Context *h){
2218 hl_decode_mb_444_internal(h, 1, 0);
2219 }
2220
2221 void ff_h264_hl_decode_mb(H264Context *h){
2222 MpegEncContext * const s = &h->s;
2223 const int mb_xy= h->mb_xy;
2224 const int mb_type = s->current_picture.f.mb_type[mb_xy];
2225 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2226
2227 if (CHROMA444) {
2228 if(is_complex || h->pixel_shift)
2229 hl_decode_mb_444_complex(h);
2230 else
2231 hl_decode_mb_444_simple(h);
2232 } else if (is_complex) {
2233 hl_decode_mb_complex(h);
2234 } else if (h->pixel_shift) {
2235 hl_decode_mb_simple_16(h);
2236 } else
2237 hl_decode_mb_simple_8(h);
2238 }
2239
2240 static int pred_weight_table(H264Context *h){
2241 MpegEncContext * const s = &h->s;
2242 int list, i;
2243 int luma_def, chroma_def;
2244
2245 h->use_weight= 0;
2246 h->use_weight_chroma= 0;
2247 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2248 if(h->sps.chroma_format_idc)
2249 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2250 luma_def = 1<<h->luma_log2_weight_denom;
2251 chroma_def = 1<<h->chroma_log2_weight_denom;
2252
2253 for(list=0; list<2; list++){
2254 h->luma_weight_flag[list] = 0;
2255 h->chroma_weight_flag[list] = 0;
2256 for(i=0; i<h->ref_count[list]; i++){
2257 int luma_weight_flag, chroma_weight_flag;
2258
2259 luma_weight_flag= get_bits1(&s->gb);
2260 if(luma_weight_flag){
2261 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2262 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2263 if( h->luma_weight[i][list][0] != luma_def
2264 || h->luma_weight[i][list][1] != 0) {
2265 h->use_weight= 1;
2266 h->luma_weight_flag[list]= 1;
2267 }
2268 }else{
2269 h->luma_weight[i][list][0]= luma_def;
2270 h->luma_weight[i][list][1]= 0;
2271 }
2272
2273 if(h->sps.chroma_format_idc){
2274 chroma_weight_flag= get_bits1(&s->gb);
2275 if(chroma_weight_flag){
2276 int j;
2277 for(j=0; j<2; j++){
2278 h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2279 h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2280 if( h->chroma_weight[i][list][j][0] != chroma_def
2281 || h->chroma_weight[i][list][j][1] != 0) {
2282 h->use_weight_chroma= 1;
2283 h->chroma_weight_flag[list]= 1;
2284 }
2285 }
2286 }else{
2287 int j;
2288 for(j=0; j<2; j++){
2289 h->chroma_weight[i][list][j][0]= chroma_def;
2290 h->chroma_weight[i][list][j][1]= 0;
2291 }
2292 }
2293 }
2294 }
2295 if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
2296 }
2297 h->use_weight= h->use_weight || h->use_weight_chroma;
2298 return 0;
2299 }
2300
2301 /**
2302 * Initialize implicit_weight table.
2303 * @param field 0/1 initialize the weight for interlaced MBAFF
2304 * -1 initializes the rest
2305 */
2306 static void implicit_weight_table(H264Context *h, int field){
2307 MpegEncContext * const s = &h->s;
2308 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2309
2310 for (i = 0; i < 2; i++) {
2311 h->luma_weight_flag[i] = 0;
2312 h->chroma_weight_flag[i] = 0;
2313 }
2314
2315 if(field < 0){
2316 if (s->picture_structure == PICT_FRAME) {
2317 cur_poc = s->current_picture_ptr->poc;
2318 } else {
2319 cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2320 }
2321 if( h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
2322 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2323 h->use_weight= 0;
2324 h->use_weight_chroma= 0;
2325 return;
2326 }
2327 ref_start= 0;
2328 ref_count0= h->ref_count[0];
2329 ref_count1= h->ref_count[1];
2330 }else{
2331 cur_poc = s->current_picture_ptr->field_poc[field];
2332 ref_start= 16;
2333 ref_count0= 16+2*h->ref_count[0];
2334 ref_count1= 16+2*h->ref_count[1];
2335 }
2336
2337 h->use_weight= 2;
2338 h->use_weight_chroma= 2;
2339 h->luma_log2_weight_denom= 5;
2340 h->chroma_log2_weight_denom= 5;
2341
2342 for(ref0=ref_start; ref0 < ref_count0; ref0++){
2343 int poc0 = h->ref_list[0][ref0].poc;
2344 for(ref1=ref_start; ref1 < ref_count1; ref1++){
2345 int w = 32;
2346 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2347 int poc1 = h->ref_list[1][ref1].poc;
2348 int td = av_clip(poc1 - poc0, -128, 127);
2349 if(td){
2350 int tb = av_clip(cur_poc - poc0, -128, 127);
2351 int tx = (16384 + (FFABS(td) >> 1)) / td;
2352 int dist_scale_factor = (tb*tx + 32) >> 8;
2353 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2354 w = 64 - dist_scale_factor;
2355 }
2356 }
2357 if(field<0){
2358 h->implicit_weight[ref0][ref1][0]=
2359 h->implicit_weight[ref0][ref1][1]= w;
2360 }else{
2361 h->implicit_weight[ref0][ref1][field]=w;
2362 }
2363 }
2364 }
2365 }
2366
2367 /**
2368 * instantaneous decoder refresh.
2369 */
2370 static void idr(H264Context *h){
2371 ff_h264_remove_all_refs(h);
2372 h->prev_frame_num= 0;
2373 h->prev_frame_num_offset= 0;
2374 h->prev_poc_msb=
2375 h->prev_poc_lsb= 0;
2376 }
2377
2378 /* forget old pics after a seek */
2379 static void flush_dpb(AVCodecContext *avctx){
2380 H264Context *h= avctx->priv_data;
2381 int i;
2382 for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
2383 if(h->delayed_pic[i])
2384 h->delayed_pic[i]->f.reference = 0;
2385 h->delayed_pic[i]= NULL;
2386 }
2387 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2388 h->last_pocs[i] = INT_MIN;
2389 h->outputed_poc=h->next_outputed_poc= INT_MIN;
2390 h->prev_interlaced_frame = 1;
2391 idr(h);
2392 if(h->s.current_picture_ptr)
2393 h->s.current_picture_ptr->f.reference = 0;
2394 h->s.first_field= 0;
2395 ff_h264_reset_sei(h);
2396 ff_mpeg_flush(avctx);
2397 }
2398
2399 static int init_poc(H264Context *h){
2400 MpegEncContext * const s = &h->s;
2401 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2402 int field_poc[2];
2403 Picture *cur = s->current_picture_ptr;
2404
2405 h->frame_num_offset= h->prev_frame_num_offset;
2406 if(h->frame_num < h->prev_frame_num)
2407 h->frame_num_offset += max_frame_num;
2408
2409 if(h->sps.poc_type==0){
2410 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2411
2412 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2413 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2414 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2415 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2416 else
2417 h->poc_msb = h->prev_poc_msb;
2418 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2419 field_poc[0] =
2420 field_poc[1] = h->poc_msb + h->poc_lsb;
2421 if(s->picture_structure == PICT_FRAME)
2422 field_poc[1] += h->delta_poc_bottom;
2423 }else if(h->sps.poc_type==1){
2424 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2425 int i;
2426
2427 if(h->sps.poc_cycle_length != 0)
2428 abs_frame_num = h->frame_num_offset + h->frame_num;
2429 else
2430 abs_frame_num = 0;
2431
2432 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2433 abs_frame_num--;
2434
2435 expected_delta_per_poc_cycle = 0;
2436 for(i=0; i < h->sps.poc_cycle_length; i++)
2437 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2438
2439 if(abs_frame_num > 0){
2440 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2441 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2442
2443 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2444 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2445 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2446 } else
2447 expectedpoc = 0;
2448
2449 if(h->nal_ref_idc == 0)
2450 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2451
2452 field_poc[0] = expectedpoc + h->delta_poc[0];
2453 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2454
2455 if(s->picture_structure == PICT_FRAME)
2456 field_poc[1] += h->delta_poc[1];
2457 }else{
2458 int poc= 2*(h->frame_num_offset + h->frame_num);
2459
2460 if(!h->nal_ref_idc)
2461 poc--;
2462
2463 field_poc[0]= poc;
2464 field_poc[1]= poc;
2465 }
2466
2467 if(s->picture_structure != PICT_BOTTOM_FIELD)
2468 s->current_picture_ptr->field_poc[0]= field_poc[0];
2469 if(s->picture_structure != PICT_TOP_FIELD)
2470 s->current_picture_ptr->field_poc[1]= field_poc[1];
2471 cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
2472
2473 return 0;
2474 }
2475
2476
2477 /**
2478 * initialize scan tables
2479 */
2480 static void init_scan_tables(H264Context *h){
2481 int i;
2482 for(i=0; i<16; i++){
2483 #define T(x) (x>>2) | ((x<<2) & 0xF)
2484 h->zigzag_scan[i] = T(zigzag_scan[i]);
2485 h-> field_scan[i] = T( field_scan[i]);
2486 #undef T
2487 }
2488 for(i=0; i<64; i++){
2489 #define T(x) (x>>3) | ((x&7)<<3)
2490 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2491 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2492 h->field_scan8x8[i] = T(field_scan8x8[i]);
2493 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
2494 #undef T
2495 }
2496 if(h->sps.transform_bypass){ //FIXME same ugly
2497 h->zigzag_scan_q0 = zigzag_scan;
2498 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
2499 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2500 h->field_scan_q0 = field_scan;
2501 h->field_scan8x8_q0 = field_scan8x8;
2502 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
2503 }else{
2504 h->zigzag_scan_q0 = h->zigzag_scan;
2505 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
2506 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2507 h->field_scan_q0 = h->field_scan;
2508 h->field_scan8x8_q0 = h->field_scan8x8;
2509 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
2510 }
2511 }
2512
2513 static int field_end(H264Context *h, int in_setup){
2514 MpegEncContext * const s = &h->s;
2515 AVCodecContext * const avctx= s->avctx;
2516 int err = 0;
2517 s->mb_y= 0;
2518
2519 if (!in_setup && !s->dropable)
2520 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
2521 s->picture_structure==PICT_BOTTOM_FIELD);
2522
2523 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2524 ff_vdpau_h264_set_reference_frames(s);
2525
2526 if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2527 if(!s->dropable) {
2528 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2529 h->prev_poc_msb= h->poc_msb;
2530 h->prev_poc_lsb= h->poc_lsb;
2531 }
2532 h->prev_frame_num_offset= h->frame_num_offset;
2533 h->prev_frame_num= h->frame_num;
2534 h->outputed_poc = h->next_outputed_poc;
2535 }
2536
2537 if (avctx->hwaccel) {
2538 if (avctx->hwaccel->end_frame(avctx) < 0)
2539 av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2540 }
2541
2542 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2543 ff_vdpau_h264_picture_complete(s);
2544
2545 /*
2546 * FIXME: Error handling code does not seem to support interlaced
2547 * when slices span multiple rows
2548 * The ff_er_add_slice calls don't work right for bottom
2549 * fields; they cause massive erroneous error concealing
2550 * Error marking covers both fields (top and bottom).
2551 * This causes a mismatched s->error_count
2552 * and a bad error table. Further, the error count goes to
2553 * INT_MAX when called for bottom field, because mb_y is
2554 * past end by one (callers fault) and resync_mb_y != 0
2555 * causes problems for the first MB line, too.
2556 */
2557 if (!FIELD_PICTURE)
2558 ff_er_frame_end(s);
2559
2560 MPV_frame_end(s);
2561
2562 h->current_slice=0;
2563
2564 return err;
2565 }
2566
2567 /**
2568 * Replicate H264 "master" context to thread contexts.
2569 */
2570 static void clone_slice(H264Context *dst, H264Context *src)
2571 {
2572 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2573 dst->s.current_picture_ptr = src->s.current_picture_ptr;
2574 dst->s.current_picture = src->s.current_picture;
2575 dst->s.linesize = src->s.linesize;
2576 dst->s.uvlinesize = src->s.uvlinesize;
2577 dst->s.first_field = src->s.first_field;
2578
2579 dst->prev_poc_msb = src->prev_poc_msb;
2580 dst->prev_poc_lsb = src->prev_poc_lsb;
2581 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2582 dst->prev_frame_num = src->prev_frame_num;
2583 dst->short_ref_count = src->short_ref_count;
2584
2585 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2586 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2587 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2588 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
2589
2590 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2591 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
2592 }
2593
2594 /**
2595 * computes profile from profile_idc and constraint_set?_flags
2596 *
2597 * @param sps SPS
2598 *
2599 * @return profile as defined by FF_PROFILE_H264_*
2600 */
2601 int ff_h264_get_profile(SPS *sps)
2602 {
2603 int profile = sps->profile_idc;
2604
2605 switch(sps->profile_idc) {
2606 case FF_PROFILE_H264_BASELINE:
2607 // constraint_set1_flag set to 1
2608 profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2609 break;
2610 case FF_PROFILE_H264_HIGH_10:
2611 case FF_PROFILE_H264_HIGH_422:
2612 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2613 // constraint_set3_flag set to 1
2614 profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2615 break;
2616 }
2617
2618 return profile;
2619 }
2620
2621 /**
2622 * decodes a slice header.
2623 * This will also call MPV_common_init() and frame_start() as needed.
2624 *
2625 * @param h h264context
2626 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2627 *
2628 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2629 */
2630 static int decode_slice_header(H264Context *h, H264Context *h0){
2631 MpegEncContext * const s = &h->s;
2632 MpegEncContext * const s0 = &h0->s;
2633 unsigned int first_mb_in_slice;
2634 unsigned int pps_id;
2635 int num_ref_idx_active_override_flag;
2636 unsigned int slice_type, tmp, i, j;
2637 int default_ref_list_done = 0;
2638 int last_pic_structure;
2639
2640 s->dropable= h->nal_ref_idc == 0;
2641
2642 /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2643 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
2644 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2645 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2646 }else{
2647 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2648 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2649 }
2650
2651 first_mb_in_slice= get_ue_golomb(&s->gb);
2652
2653 if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2654 if(h0->current_slice && FIELD_PICTURE){
2655 field_end(h, 1);
2656 }
2657
2658 h0->current_slice = 0;
2659 if (!s0->first_field)
2660 s->current_picture_ptr= NULL;
2661 }
2662
2663 slice_type= get_ue_golomb_31(&s->gb);
2664 if(slice_type > 9){
2665 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);
2666 return -1;
2667 }
2668 if(slice_type > 4){
2669 slice_type -= 5;
2670 h->slice_type_fixed=1;
2671 }else
2672 h->slice_type_fixed=0;
2673
2674 slice_type= golomb_to_pict_type[ slice_type ];
2675 if (slice_type == AV_PICTURE_TYPE_I
2676 || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
2677 default_ref_list_done = 1;
2678 }
2679 h->slice_type= slice_type;
2680 h->slice_type_nos= slice_type & 3;
2681
2682 s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
2683
2684 pps_id= get_ue_golomb(&s->gb);
2685 if(pps_id>=MAX_PPS_COUNT){
2686 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2687 return -1;
2688 }
2689 if(!h0->pps_buffers[pps_id]) {
2690 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
2691 return -1;
2692 }
2693 h->pps= *h0->pps_buffers[pps_id];
2694
2695 if(!h0->sps_buffers[h->pps.sps_id]) {
2696 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
2697 return -1;
2698 }
2699 h->sps = *h0->sps_buffers[h->pps.sps_id];
2700
2701 s->avctx->profile = ff_h264_get_profile(&h->sps);
2702 s->avctx->level = h->sps.level_idc;
2703 s->avctx->refs = h->sps.ref_frame_count;
2704
2705 if(h == h0 && h->dequant_coeff_pps != pps_id){
2706 h->dequant_coeff_pps = pps_id;
2707 init_dequant_tables(h);
2708 }
2709
2710 s->mb_width= h->sps.mb_width;
2711 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2712
2713 h->b_stride= s->mb_width*4;
2714
2715 s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2716
2717 s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2718 if(h->sps.frame_mbs_only_flag)
2719 s->height= 16*s->mb_height - (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2720 else
2721 s->height= 16*s->mb_height - (2<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2722
2723 if (s->context_initialized
2724 && ( s->width != s->avctx->width || s->height != s->avctx->height
2725 || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
2726 if(h != h0) {
2727 av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
2728 return -1; // width / height changed during parallelized decoding
2729 }
2730 free_tables(h, 0);
2731 flush_dpb(s->avctx);
2732 MPV_common_end(s);
2733 }
2734 if (!s->context_initialized) {
2735 if (h != h0) {
2736 av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2737 return -1;
2738 }
2739
2740 avcodec_set_dimensions(s->avctx, s->width, s->height);
2741 s->avctx->sample_aspect_ratio= h->sps.sar;
2742 av_assert0(s->avctx->sample_aspect_ratio.den);
2743
2744 if(h->sps.video_signal_type_present_flag){
2745 s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2746 if(h->sps.colour_description_present_flag){
2747 s->avctx->color_primaries = h->sps.color_primaries;
2748 s->avctx->color_trc = h->sps.color_trc;
2749 s->avctx->colorspace = h->sps.colorspace;
2750 }
2751 }
2752
2753 if(h->sps.timing_info_present_flag){
2754 int64_t den= h->sps.time_scale;
2755 if(h->x264_build < 44U)
2756 den *= 2;
2757 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2758 h->sps.num_units_in_tick, den, 1<<30);
2759 }
2760
2761 switch (h->sps.bit_depth_luma) {
2762 case 9 :
2763 if (CHROMA444) {
2764 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2765 s->avctx->pix_fmt = PIX_FMT_GBRP9;
2766 } else
2767 s->avctx->pix_fmt = PIX_FMT_YUV444P9;
2768 } else if (CHROMA422)
2769 s->avctx->pix_fmt = PIX_FMT_YUV422P9;
2770 else
2771 s->avctx->pix_fmt = PIX_FMT_YUV420P9;
2772 break;
2773 case 10 :
2774 if (CHROMA444) {
2775 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2776 s->avctx->pix_fmt = PIX_FMT_GBRP10;
2777 } else
2778 s->avctx->pix_fmt = PIX_FMT_YUV444P10;
2779 } else if (CHROMA422)
2780 s->avctx->pix_fmt = PIX_FMT_YUV422P10;
2781 else
2782 s->avctx->pix_fmt = PIX_FMT_YUV420P10;
2783 break;
2784 default:
2785 if (CHROMA444){
2786 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2787 s->avctx->pix_fmt = PIX_FMT_GBRP;
2788 } else
2789 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
2790 } else if (CHROMA422) {
2791 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P : PIX_FMT_YUV422P;
2792 }else{
2793 s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2794 s->avctx->codec->pix_fmts ?
2795 s->avctx->codec->pix_fmts :
2796 s->avctx->color_range == AVCOL_RANGE_JPEG ?
2797 hwaccel_pixfmt_list_h264_jpeg_420 :
2798 ff_hwaccel_pixfmt_list_420);
2799 }
2800 }
2801
2802 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2803
2804 if (MPV_common_init(s) < 0) {
2805 av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed.\n");
2806 return -1;
2807 }
2808 s->first_field = 0;
2809 h->prev_interlaced_frame = 1;
2810
2811 init_scan_tables(h);
2812 if (ff_h264_alloc_tables(h) < 0) {
2813 av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2814 return AVERROR(ENOMEM);
2815 }
2816
2817 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
2818 if (context_init(h) < 0) {
2819 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2820 return -1;
2821 }
2822 } else {
2823 for(i = 1; i < s->avctx->thread_count; i++) {
2824 H264Context *c;
2825 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2826 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2827 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2828 c->h264dsp = h->h264dsp;
2829 c->sps = h->sps;
2830 c->pps = h->pps;
2831 c->pixel_shift = h->pixel_shift;
2832 init_scan_tables(c);
2833 clone_tables(c, h, i);
2834 }
2835
2836 for(i = 0; i < s->avctx->thread_count; i++)
2837 if (context_init(h->thread_context[i]) < 0) {
2838 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2839 return -1;
2840 }
2841 }
2842 }
2843
2844 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2845
2846 h->mb_mbaff = 0;
2847 h->mb_aff_frame = 0;
2848 last_pic_structure = s0->picture_structure;
2849 if(h->sps.frame_mbs_only_flag){
2850 s->picture_structure= PICT_FRAME;
2851 }else{
2852 if(get_bits1(&s->gb)) { //field_pic_flag
2853 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2854 } else {
2855 s->picture_structure= PICT_FRAME;
2856 h->mb_aff_frame = h->sps.mb_aff;
2857 }
2858 }
2859 h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2860
2861 if(h0->current_slice == 0){
2862 // Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
2863 if(h->frame_num != h->prev_frame_num) {
2864 int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
2865
2866 if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2867
2868 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2869 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2870 if (unwrap_prev_frame_num < 0)
2871 unwrap_prev_frame_num += max_frame_num;
2872
2873 h->prev_frame_num = unwrap_prev_frame_num;
2874 }
2875 }
2876
2877 while(h->frame_num != h->prev_frame_num &&
2878 h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
2879 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2880 av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
2881 if (ff_h264_frame_start(h) < 0)
2882 return -1;
2883 h->prev_frame_num++;
2884 h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2885 s->current_picture_ptr->frame_num= h->prev_frame_num;
2886 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
2887 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
2888 ff_generate_sliding_window_mmcos(h);
2889 if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2890 (s->avctx->err_recognition & AV_EF_EXPLODE))
2891 return AVERROR_INVALIDDATA;
2892 /* Error concealment: if a ref is missing, copy the previous ref in its place.
2893 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2894 * about there being no actual duplicates.
2895 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
2896 * concealing a lost frame, this probably isn't noticable by comparison, but it should
2897 * be fixed. */
2898 if (h->short_ref_count) {
2899 if (prev) {
2900 av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2901 (const uint8_t**)prev->f.data, prev->f.linesize,
2902 s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2903 h->short_ref[0]->poc = prev->poc+2;
2904 }
2905 h->short_ref[0]->frame_num = h->prev_frame_num;
2906 }
2907 }
2908
2909 /* See if we have a decoded first field looking for a pair... */
2910 if (s0->first_field) {
2911 assert(s0->current_picture_ptr);
2912 assert(s0->current_picture_ptr->f.data[0]);
2913 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2914
2915 /* figure out if we have a complementary field pair */
2916 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2917 /*
2918 * Previous field is unmatched. Don't display it, but let it
2919 * remain for reference if marked as such.
2920 */
2921 s0->current_picture_ptr = NULL;
2922 s0->first_field = FIELD_PICTURE;
2923
2924 } else {
2925 if (h->nal_ref_idc &&
2926 s0->current_picture_ptr->f.reference &&
2927 s0->current_picture_ptr->frame_num != h->frame_num) {
2928 /*
2929 * This and previous field were reference, but had
2930 * different frame_nums. Consider this field first in
2931 * pair. Throw away previous field except for reference
2932 * purposes.
2933 */
2934 s0->first_field = 1;
2935 s0->current_picture_ptr = NULL;
2936
2937 } else {
2938 /* Second field in complementary pair */
2939 s0->first_field = 0;
2940 }
2941 }
2942
2943 } else {
2944 /* Frame or first field in a potentially complementary pair */
2945 assert(!s0->current_picture_ptr);
2946 s0->first_field = FIELD_PICTURE;
2947 }
2948
2949 if(!FIELD_PICTURE || s0->first_field) {
2950 if (ff_h264_frame_start(h) < 0) {
2951 s0->first_field = 0;
2952 return -1;
2953 }
2954 } else {
2955 ff_release_unused_pictures(s, 0);
2956 }
2957 }
2958 if(h != h0)
2959 clone_slice(h, h0);
2960
2961 s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2962
2963 assert(s->mb_num == s->mb_width * s->mb_height);
2964 if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2965 first_mb_in_slice >= s->mb_num){
2966 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2967 return -1;
2968 }
2969 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2970 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2971 if (s->picture_structure == PICT_BOTTOM_FIELD)
2972 s->resync_mb_y = s->mb_y = s->mb_y + 1;
2973 assert(s->mb_y < s->mb_height);
2974
2975 if(s->picture_structure==PICT_FRAME){
2976 h->curr_pic_num= h->frame_num;
2977 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2978 }else{
2979 h->curr_pic_num= 2*h->frame_num + 1;
2980 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2981 }
2982
2983 if(h->nal_unit_type == NAL_IDR_SLICE){
2984 get_ue_golomb(&s->gb); /* idr_pic_id */
2985 }
2986
2987 if(h->sps.poc_type==0){
2988 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2989
2990 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2991 h->delta_poc_bottom= get_se_golomb(&s->gb);
2992 }
2993 }
2994
2995 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2996 h->delta_poc[0]= get_se_golomb(&s->gb);
2997
2998 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2999 h->delta_poc[1]= get_se_golomb(&s->gb);
3000 }
3001
3002 init_poc(h);
3003
3004 if(h->pps.redundant_pic_cnt_present){
3005 h->redundant_pic_count= get_ue_golomb(&s->gb);
3006 }
3007
3008 //set defaults, might be overridden a few lines later
3009 h->ref_count[0]= h->pps.ref_count[0];
3010 h->ref_count[1]= h->pps.ref_count[1];
3011
3012 if(h->slice_type_nos != AV_PICTURE_TYPE_I){
3013 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
3014 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3015 }
3016 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3017
3018 if(num_ref_idx_active_override_flag){
3019 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3020 if(h->slice_type_nos==AV_PICTURE_TYPE_B)
3021 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3022
3023 if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
3024 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3025 h->ref_count[0]= h->ref_count[1]= 1;
3026 return -1;
3027 }
3028 }
3029 if(h->slice_type_nos == AV_PICTURE_TYPE_B)
3030 h->list_count= 2;
3031 else
3032 h->list_count= 1;
3033 }else
3034 h->list_count= 0;
3035
3036 if(!default_ref_list_done){
3037 ff_h264_fill_default_ref_list(h);
3038 }
3039
3040 if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0) {
3041 h->ref_count[1]= h->ref_count[0]= 0;
3042 return -1;
3043 }
3044
3045 if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
3046 s->last_picture_ptr= &h->ref_list[0][0];
3047 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3048 }
3049 if(h->slice_type_nos==AV_PICTURE_TYPE_B){
3050 s->next_picture_ptr= &h->ref_list[1][0];
3051 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3052 }
3053
3054 if( (h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P )
3055 || (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
3056 pred_weight_table(h);
3057 else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
3058 implicit_weight_table(h, -1);
3059 }else {
3060 h->use_weight = 0;
3061 for (i = 0; i < 2; i++) {
3062 h->luma_weight_flag[i] = 0;
3063 h->chroma_weight_flag[i] = 0;
3064 }
3065 }
3066
3067 if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
3068 (s->avctx->err_recognition & AV_EF_EXPLODE))
3069 return AVERROR_INVALIDDATA;
3070
3071 if(FRAME_MBAFF){
3072 ff_h264_fill_mbaff_ref_list(h);
3073
3074 if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
3075 implicit_weight_table(h, 0);
3076 implicit_weight_table(h, 1);
3077 }
3078 }
3079
3080 if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3081 ff_h264_direct_dist_scale_factor(h);
3082 ff_h264_direct_ref_list_init(h);
3083
3084 if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
3085 tmp = get_ue_golomb_31(&s->gb);
3086 if(tmp > 2){
3087 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3088 return -1;
3089 }
3090 h->cabac_init_idc= tmp;
3091 }
3092
3093 h->last_qscale_diff = 0;
3094 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3095 if(tmp>51+6*(h->sps.bit_depth_luma-8)){
3096 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3097 return -1;
3098 }