77acd7168f0342d6e3cfc48ae46a18223381b44f
[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 * Check if the top & left blocks are available if needed and
66 * change the dc mode so it only uses the available blocks.
67 */
68 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
69 MpegEncContext * const s = &h->s;
70 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
71 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
72 int i;
73
74 if(!(h->top_samples_available&0x8000)){
75 for(i=0; i<4; i++){
76 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
77 if(status<0){
78 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);
79 return -1;
80 } else if(status){
81 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
82 }
83 }
84 }
85
86 if((h->left_samples_available&0x8888)!=0x8888){
87 static const int mask[4]={0x8000,0x2000,0x80,0x20};
88 for(i=0; i<4; i++){
89 if(!(h->left_samples_available&mask[i])){
90 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
91 if(status<0){
92 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);
93 return -1;
94 } else if(status){
95 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
96 }
97 }
98 }
99 }
100
101 return 0;
102 } //FIXME cleanup like ff_h264_check_intra_pred_mode
103
104 /**
105 * Check if the top & left blocks are available if needed and
106 * change the dc mode so it only uses the available blocks.
107 */
108 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
109 MpegEncContext * const s = &h->s;
110 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
111 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
112
113 if(mode > 6U) {
114 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
115 return -1;
116 }
117
118 if(!(h->top_samples_available&0x8000)){
119 mode= top[ mode ];
120 if(mode<0){
121 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
122 return -1;
123 }
124 }
125
126 if((h->left_samples_available&0x8080) != 0x8080){
127 mode= left[ mode ];
128 if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
129 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
130 }
131 if(mode<0){
132 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
133 return -1;
134 }
135 }
136
137 return mode;
138 }
139
140 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
141 int i, si, di;
142 uint8_t *dst;
143 int bufidx;
144
145 // src[0]&0x80; //forbidden bit
146 h->nal_ref_idc= src[0]>>5;
147 h->nal_unit_type= src[0]&0x1F;
148
149 src++; length--;
150
151 #if HAVE_FAST_UNALIGNED
152 # if HAVE_FAST_64BIT
153 # define RS 7
154 for(i=0; i+1<length; i+=9){
155 if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
156 # else
157 # define RS 3
158 for(i=0; i+1<length; i+=5){
159 if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
160 # endif
161 continue;
162 if(i>0 && !src[i]) i--;
163 while(src[i]) i++;
164 #else
165 # define RS 0
166 for(i=0; i+1<length; i+=2){
167 if(src[i]) continue;
168 if(i>0 && src[i-1]==0) i--;
169 #endif
170 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
171 if(src[i+2]!=3){
172 /* startcode, so we must be past the end */
173 length=i;
174 }
175 break;
176 }
177 i-= RS;
178 }
179
180 if(i>=length-1){ //no escaped 0
181 *dst_length= length;
182 *consumed= length+1; //+1 for the header
183 return src;
184 }
185
186 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
187 av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
188 dst= h->rbsp_buffer[bufidx];
189
190 if (dst == NULL){
191 return NULL;
192 }
193
194 //printf("decoding esc\n");
195 memcpy(dst, src, i);
196 si=di=i;
197 while(si+2<length){
198 //remove escapes (very rare 1:2^22)
199 if(src[si+2]>3){
200 dst[di++]= src[si++];
201 dst[di++]= src[si++];
202 }else if(src[si]==0 && src[si+1]==0){
203 if(src[si+2]==3){ //escape
204 dst[di++]= 0;
205 dst[di++]= 0;
206 si+=3;
207 continue;
208 }else //next start code
209 goto nsc;
210 }
211
212 dst[di++]= src[si++];
213 }
214 while(si<length)
215 dst[di++]= src[si++];
216 nsc:
217
218 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
219
220 *dst_length= di;
221 *consumed= si + 1;//+1 for the header
222 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
223 return dst;
224 }
225
226 /**
227 * Identify the exact end of the bitstream
228 * @return the length of the trailing, or 0 if damaged
229 */
230 static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
231 int v= *src;
232 int r;
233
234 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
235
236 for(r=1; r<9; r++){
237 if(v&1) return r;
238 v>>=1;
239 }
240 return 0;
241 }
242
243 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height,
244 int y_offset, int list){
245 int raw_my= h->mv_cache[list][ scan8[n] ][1];
246 int filter_height= (raw_my&3) ? 2 : 0;
247 int full_my= (raw_my>>2) + y_offset;
248 int top = full_my - filter_height, bottom = full_my + height + filter_height;
249
250 return FFMAX(abs(top), bottom);
251 }
252
253 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height,
254 int y_offset, int list0, int list1, int *nrefs){
255 MpegEncContext * const s = &h->s;
256 int my;
257
258 y_offset += 16*(s->mb_y >> MB_FIELD);
259
260 if(list0){
261 int ref_n = h->ref_cache[0][ scan8[n] ];
262 Picture *ref= &h->ref_list[0][ref_n];
263
264 // Error resilience puts the current picture in the ref list.
265 // Don't try to wait on these as it will cause a deadlock.
266 // Fields can wait on each other, though.
267 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
268 (ref->f.reference & 3) != s->picture_structure) {
269 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
270 if (refs[0][ref_n] < 0) nrefs[0] += 1;
271 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
272 }
273 }
274
275 if(list1){
276 int ref_n = h->ref_cache[1][ scan8[n] ];
277 Picture *ref= &h->ref_list[1][ref_n];
278
279 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
280 (ref->f.reference & 3) != s->picture_structure) {
281 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
282 if (refs[1][ref_n] < 0) nrefs[1] += 1;
283 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
284 }
285 }
286 }
287
288 /**
289 * Wait until all reference frames are available for MC operations.
290 *
291 * @param h the H264 context
292 */
293 static void await_references(H264Context *h){
294 MpegEncContext * const s = &h->s;
295 const int mb_xy= h->mb_xy;
296 const int mb_type = s->current_picture.f.mb_type[mb_xy];
297 int refs[2][48];
298 int nrefs[2] = {0};
299 int ref, list;
300
301 memset(refs, -1, sizeof(refs));
302
303 if(IS_16X16(mb_type)){
304 get_lowest_part_y(h, refs, 0, 16, 0,
305 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
306 }else if(IS_16X8(mb_type)){
307 get_lowest_part_y(h, refs, 0, 8, 0,
308 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
309 get_lowest_part_y(h, refs, 8, 8, 8,
310 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
311 }else if(IS_8X16(mb_type)){
312 get_lowest_part_y(h, refs, 0, 16, 0,
313 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
314 get_lowest_part_y(h, refs, 4, 16, 0,
315 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
316 }else{
317 int i;
318
319 assert(IS_8X8(mb_type));
320
321 for(i=0; i<4; i++){
322 const int sub_mb_type= h->sub_mb_type[i];
323 const int n= 4*i;
324 int y_offset= (i&2)<<2;
325
326 if(IS_SUB_8X8(sub_mb_type)){
327 get_lowest_part_y(h, refs, n , 8, y_offset,
328 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
329 }else if(IS_SUB_8X4(sub_mb_type)){
330 get_lowest_part_y(h, refs, n , 4, y_offset,
331 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
332 get_lowest_part_y(h, refs, n+2, 4, y_offset+4,
333 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
334 }else if(IS_SUB_4X8(sub_mb_type)){
335 get_lowest_part_y(h, refs, n , 8, y_offset,
336 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
337 get_lowest_part_y(h, refs, n+1, 8, y_offset,
338 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
339 }else{
340 int j;
341 assert(IS_SUB_4X4(sub_mb_type));
342 for(j=0; j<4; j++){
343 int sub_y_offset= y_offset + 2*(j&2);
344 get_lowest_part_y(h, refs, n+j, 4, sub_y_offset,
345 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
346 }
347 }
348 }
349 }
350
351 for(list=h->list_count-1; list>=0; list--){
352 for(ref=0; ref<48 && nrefs[list]; ref++){
353 int row = refs[list][ref];
354 if(row >= 0){
355 Picture *ref_pic = &h->ref_list[list][ref];
356 int ref_field = ref_pic->f.reference - 1;
357 int ref_field_picture = ref_pic->field_picture;
358 int pic_height = 16*s->mb_height >> ref_field_picture;
359
360 row <<= MB_MBAFF;
361 nrefs[list]--;
362
363 if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields
364 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) - !(row&1), pic_height-1), 1);
365 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) , pic_height-1), 0);
366 }else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame
367 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row*2 + ref_field , pic_height-1), 0);
368 }else if(FIELD_PICTURE){
369 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), ref_field);
370 }else{
371 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), 0);
372 }
373 }
374 }
375 }
376 }
377
378 #if 0
379 /**
380 * DCT transforms the 16 dc values.
381 * @param qp quantization parameter ??? FIXME
382 */
383 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
384 // const int qmul= dequant_coeff[qp][0];
385 int i;
386 int temp[16]; //FIXME check if this is a good idea
387 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
388 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
389
390 for(i=0; i<4; i++){
391 const int offset= y_offset[i];
392 const int z0= block[offset+stride*0] + block[offset+stride*4];
393 const int z1= block[offset+stride*0] - block[offset+stride*4];
394 const int z2= block[offset+stride*1] - block[offset+stride*5];
395 const int z3= block[offset+stride*1] + block[offset+stride*5];
396
397 temp[4*i+0]= z0+z3;
398 temp[4*i+1]= z1+z2;
399 temp[4*i+2]= z1-z2;
400 temp[4*i+3]= z0-z3;
401 }
402
403 for(i=0; i<4; i++){
404 const int offset= x_offset[i];
405 const int z0= temp[4*0+i] + temp[4*2+i];
406 const int z1= temp[4*0+i] - temp[4*2+i];
407 const int z2= temp[4*1+i] - temp[4*3+i];
408 const int z3= temp[4*1+i] + temp[4*3+i];
409
410 block[stride*0 +offset]= (z0 + z3)>>1;
411 block[stride*2 +offset]= (z1 + z2)>>1;
412 block[stride*8 +offset]= (z1 - z2)>>1;
413 block[stride*10+offset]= (z0 - z3)>>1;
414 }
415 }
416 #endif
417
418 #undef xStride
419 #undef stride
420
421 #if 0
422 static void chroma_dc_dct_c(DCTELEM *block){
423 const int stride= 16*2;
424 const int xStride= 16;
425 int a,b,c,d,e;
426
427 a= block[stride*0 + xStride*0];
428 b= block[stride*0 + xStride*1];
429 c= block[stride*1 + xStride*0];
430 d= block[stride*1 + xStride*1];
431
432 e= a-b;
433 a= a+b;
434 b= c-d;
435 c= c+d;
436
437 block[stride*0 + xStride*0]= (a+c);
438 block[stride*0 + xStride*1]= (e+b);
439 block[stride*1 + xStride*0]= (a-c);
440 block[stride*1 + xStride*1]= (e-b);
441 }
442 #endif
443
444 static av_always_inline void
445 mc_dir_part(H264Context *h, Picture *pic, int n, int square,
446 int height, int delta, int list,
447 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
448 int src_x_offset, int src_y_offset,
449 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
450 int pixel_shift, int chroma_idc)
451 {
452 MpegEncContext * const s = &h->s;
453 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
454 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
455 const int luma_xy= (mx&3) + ((my&3)<<2);
456 int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
457 uint8_t * src_y = pic->f.data[0] + offset;
458 uint8_t * src_cb, * src_cr;
459 int extra_width= h->emu_edge_width;
460 int extra_height= h->emu_edge_height;
461 int emu=0;
462 const int full_mx= mx>>2;
463 const int full_my= my>>2;
464 const int pic_width = 16*s->mb_width;
465 const int pic_height = 16*s->mb_height >> MB_FIELD;
466 int ysh;
467
468 if(mx&7) extra_width -= 3;
469 if(my&7) extra_height -= 3;
470
471 if( full_mx < 0-extra_width
472 || full_my < 0-extra_height
473 || full_mx + 16/*FIXME*/ > pic_width + extra_width
474 || full_my + 16/*FIXME*/ > pic_height + extra_height){
475 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
476 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
477 src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
478 emu=1;
479 }
480
481 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
482 if(!square){
483 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
484 }
485
486 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
487
488 if(chroma_idc == 3 /* yuv444 */){
489 src_cb = pic->f.data[1] + offset;
490 if(emu){
491 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
492 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
493 src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
494 }
495 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
496 if(!square){
497 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
498 }
499
500 src_cr = pic->f.data[2] + offset;
501 if(emu){
502 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
503 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
504 src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
505 }
506 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
507 if(!square){
508 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
509 }
510 return;
511 }
512
513 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
514 if(chroma_idc == 1 /* yuv420 */ && MB_FIELD){
515 // chroma offset when predicting from a field of opposite parity
516 my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
517 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
518 }
519
520 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
521 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
522
523 if(emu){
524 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
525 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
526 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
527 src_cb= s->edge_emu_buffer;
528 }
529 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
530 mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
531
532 if(emu){
533 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
534 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
535 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
536 src_cr= s->edge_emu_buffer;
537 }
538 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
539 mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
540 }
541
542 static av_always_inline void
543 mc_part_std(H264Context *h, int n, int square, int height, int delta,
544 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
545 int x_offset, int y_offset,
546 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
547 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
548 int list0, int list1, int pixel_shift, int chroma_idc)
549 {
550 MpegEncContext * const s = &h->s;
551 qpel_mc_func *qpix_op= qpix_put;
552 h264_chroma_mc_func chroma_op= chroma_put;
553
554 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
555 if (chroma_idc == 3 /* yuv444 */) {
556 dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
557 dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
558 } else if (chroma_idc == 2 /* yuv422 */) {
559 dest_cb += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
560 dest_cr += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
561 } else /* yuv420 */ {
562 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
563 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
564 }
565 x_offset += 8*s->mb_x;
566 y_offset += 8*(s->mb_y >> MB_FIELD);
567
568 if(list0){
569 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
570 mc_dir_part(h, ref, n, square, height, delta, 0,
571 dest_y, dest_cb, dest_cr, x_offset, y_offset,
572 qpix_op, chroma_op, pixel_shift, chroma_idc);
573
574 qpix_op= qpix_avg;
575 chroma_op= chroma_avg;
576 }
577
578 if(list1){
579 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
580 mc_dir_part(h, ref, n, square, height, delta, 1,
581 dest_y, dest_cb, dest_cr, x_offset, y_offset,
582 qpix_op, chroma_op, pixel_shift, chroma_idc);
583 }
584 }
585
586 static av_always_inline void
587 mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
588 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
589 int x_offset, int y_offset,
590 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
591 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
592 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
593 int list0, int list1, int pixel_shift, int chroma_idc){
594 MpegEncContext * const s = &h->s;
595 int chroma_height;
596
597 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
598 if (chroma_idc == 3 /* yuv444 */) {
599 chroma_height = height;
600 chroma_weight_avg = luma_weight_avg;
601 chroma_weight_op = luma_weight_op;
602 dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
603 dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
604 } else if (chroma_idc == 2 /* yuv422 */) {
605 chroma_height = height;
606 dest_cb += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
607 dest_cr += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
608 } else /* yuv420 */ {
609 chroma_height = height >> 1;
610 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
611 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
612 }
613 x_offset += 8*s->mb_x;
614 y_offset += 8*(s->mb_y >> MB_FIELD);
615
616 if(list0 && list1){
617 /* don't optimize for luma-only case, since B-frames usually
618 * use implicit weights => chroma too. */
619 uint8_t *tmp_cb = s->obmc_scratchpad;
620 uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
621 uint8_t *tmp_y = s->obmc_scratchpad + 16*h->mb_uvlinesize;
622 int refn0 = h->ref_cache[0][ scan8[n] ];
623 int refn1 = h->ref_cache[1][ scan8[n] ];
624
625 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
626 dest_y, dest_cb, dest_cr,
627 x_offset, y_offset, qpix_put, chroma_put,
628 pixel_shift, chroma_idc);
629 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
630 tmp_y, tmp_cb, tmp_cr,
631 x_offset, y_offset, qpix_put, chroma_put,
632 pixel_shift, chroma_idc);
633
634 if(h->use_weight == 2){
635 int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
636 int weight1 = 64 - weight0;
637 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize,
638 height, 5, weight0, weight1, 0);
639 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
640 chroma_height, 5, weight0, weight1, 0);
641 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
642 chroma_height, 5, weight0, weight1, 0);
643 }else{
644 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height, h->luma_log2_weight_denom,
645 h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
646 h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
647 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
648 h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
649 h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
650 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
651 h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
652 h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
653 }
654 }else{
655 int list = list1 ? 1 : 0;
656 int refn = h->ref_cache[list][ scan8[n] ];
657 Picture *ref= &h->ref_list[list][refn];
658 mc_dir_part(h, ref, n, square, height, delta, list,
659 dest_y, dest_cb, dest_cr, x_offset, y_offset,
660 qpix_put, chroma_put, pixel_shift, chroma_idc);
661
662 luma_weight_op(dest_y, h->mb_linesize, height, h->luma_log2_weight_denom,
663 h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
664 if(h->use_weight_chroma){
665 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
666 h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
667 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
668 h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
669 }
670 }
671 }
672
673 static av_always_inline void
674 mc_part(H264Context *h, int n, int square, int height, int delta,
675 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
676 int x_offset, int y_offset,
677 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
678 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
679 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
680 int list0, int list1, int pixel_shift, int chroma_idc)
681 {
682 if((h->use_weight==2 && list0 && list1
683 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
684 || h->use_weight==1)
685 mc_part_weighted(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
686 x_offset, y_offset, qpix_put, chroma_put,
687 weight_op[0], weight_op[1], weight_avg[0],
688 weight_avg[1], list0, list1, pixel_shift, chroma_idc);
689 else
690 mc_part_std(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
691 x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
692 chroma_avg, list0, list1, pixel_shift, chroma_idc);
693 }
694
695 static av_always_inline void
696 prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc)
697 {
698 /* fetch pixels for estimated mv 4 macroblocks ahead
699 * optimized for 64byte cache lines */
700 MpegEncContext * const s = &h->s;
701 const int refn = h->ref_cache[list][scan8[0]];
702 if(refn >= 0){
703 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
704 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
705 uint8_t **src = h->ref_list[list][refn].f.data;
706 int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
707 s->dsp.prefetch(src[0]+off, s->linesize, 4);
708 if (chroma_idc == 3 /* yuv444 */) {
709 s->dsp.prefetch(src[1]+off, s->linesize, 4);
710 s->dsp.prefetch(src[2]+off, s->linesize, 4);
711 }else{
712 off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
713 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
714 }
715 }
716 }
717
718 static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
719 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
720 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
721 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
722 int pixel_shift, int chroma_idc)
723 {
724 MpegEncContext * const s = &h->s;
725 const int mb_xy= h->mb_xy;
726 const int mb_type = s->current_picture.f.mb_type[mb_xy];
727
728 assert(IS_INTER(mb_type));
729
730 if(HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
731 await_references(h);
732 prefetch_motion(h, 0, pixel_shift, chroma_idc);
733
734 if(IS_16X16(mb_type)){
735 mc_part(h, 0, 1, 16, 0, dest_y, dest_cb, dest_cr, 0, 0,
736 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
737 weight_op, weight_avg,
738 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
739 pixel_shift, chroma_idc);
740 }else if(IS_16X8(mb_type)){
741 mc_part(h, 0, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
742 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
743 weight_op, weight_avg,
744 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
745 pixel_shift, chroma_idc);
746 mc_part(h, 8, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
747 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
748 weight_op, weight_avg,
749 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
750 pixel_shift, chroma_idc);
751 }else if(IS_8X16(mb_type)){
752 mc_part(h, 0, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
753 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
754 &weight_op[1], &weight_avg[1],
755 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
756 pixel_shift, chroma_idc);
757 mc_part(h, 4, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
758 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
759 &weight_op[1], &weight_avg[1],
760 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
761 pixel_shift, chroma_idc);
762 }else{
763 int i;
764
765 assert(IS_8X8(mb_type));
766
767 for(i=0; i<4; i++){
768 const int sub_mb_type= h->sub_mb_type[i];
769 const int n= 4*i;
770 int x_offset= (i&1)<<2;
771 int y_offset= (i&2)<<1;
772
773 if(IS_SUB_8X8(sub_mb_type)){
774 mc_part(h, n, 1, 8, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
775 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
776 &weight_op[1], &weight_avg[1],
777 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
778 pixel_shift, chroma_idc);
779 }else if(IS_SUB_8X4(sub_mb_type)){
780 mc_part(h, n , 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
781 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
782 &weight_op[1], &weight_avg[1],
783 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
784 pixel_shift, chroma_idc);
785 mc_part(h, n+2, 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
786 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
787 &weight_op[1], &weight_avg[1],
788 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
789 pixel_shift, chroma_idc);
790 }else if(IS_SUB_4X8(sub_mb_type)){
791 mc_part(h, n , 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
792 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
793 &weight_op[2], &weight_avg[2],
794 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
795 pixel_shift, chroma_idc);
796 mc_part(h, n+1, 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
797 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
798 &weight_op[2], &weight_avg[2],
799 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
800 pixel_shift, chroma_idc);
801 }else{
802 int j;
803 assert(IS_SUB_4X4(sub_mb_type));
804 for(j=0; j<4; j++){
805 int sub_x_offset= x_offset + 2*(j&1);
806 int sub_y_offset= y_offset + (j&2);
807 mc_part(h, n+j, 1, 4, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
808 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
809 &weight_op[2], &weight_avg[2],
810 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
811 pixel_shift, chroma_idc);
812 }
813 }
814 }
815 }
816
817 prefetch_motion(h, 1, pixel_shift, chroma_idc);
818 }
819
820 static av_always_inline void
821 hl_motion_420(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
822 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
823 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
824 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
825 int pixel_shift)
826 {
827 hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
828 qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 1);
829 }
830
831 static av_always_inline void
832 hl_motion_422(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
833 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
834 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
835 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
836 int pixel_shift)
837 {
838 hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
839 qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 2);
840 }
841
842 static void free_tables(H264Context *h, int free_rbsp){
843 int i;
844 H264Context *hx;
845
846 av_freep(&h->intra4x4_pred_mode);
847 av_freep(&h->chroma_pred_mode_table);
848 av_freep(&h->cbp_table);
849 av_freep(&h->mvd_table[0]);
850 av_freep(&h->mvd_table[1]);
851 av_freep(&h->direct_table);
852 av_freep(&h->non_zero_count);
853 av_freep(&h->slice_table_base);
854 h->slice_table= NULL;
855 av_freep(&h->list_counts);
856
857 av_freep(&h->mb2b_xy);
858 av_freep(&h->mb2br_xy);
859
860 for(i = 0; i < MAX_THREADS; i++) {
861 hx = h->thread_context[i];
862 if(!hx) continue;
863 av_freep(&hx->top_borders[1]);
864 av_freep(&hx->top_borders[0]);
865 av_freep(&hx->s.obmc_scratchpad);
866 if (free_rbsp){
867 av_freep(&hx->rbsp_buffer[1]);
868 av_freep(&hx->rbsp_buffer[0]);
869 hx->rbsp_buffer_size[0] = 0;
870 hx->rbsp_buffer_size[1] = 0;
871 }
872 if (i) av_freep(&h->thread_context[i]);
873 }
874 }
875
876 static void init_dequant8_coeff_table(H264Context *h){
877 int i,j,q,x;
878 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
879
880 for(i=0; i<6; i++ ){
881 h->dequant8_coeff[i] = h->dequant8_buffer[i];
882 for(j=0; j<i; j++){
883 if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
884 h->dequant8_coeff[i] = h->dequant8_buffer[j];
885 break;
886 }
887 }
888 if(j<i)
889 continue;
890
891 for(q=0; q<max_qp+1; q++){
892 int shift = div6[q];
893 int idx = rem6[q];
894 for(x=0; x<64; x++)
895 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
896 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
897 h->pps.scaling_matrix8[i][x]) << shift;
898 }
899 }
900 }
901
902 static void init_dequant4_coeff_table(H264Context *h){
903 int i,j,q,x;
904 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
905 for(i=0; i<6; i++ ){
906 h->dequant4_coeff[i] = h->dequant4_buffer[i];
907 for(j=0; j<i; j++){
908 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
909 h->dequant4_coeff[i] = h->dequant4_buffer[j];
910 break;
911 }
912 }
913 if(j<i)
914 continue;
915
916 for(q=0; q<max_qp+1; q++){
917 int shift = div6[q] + 2;
918 int idx = rem6[q];
919 for(x=0; x<16; x++)
920 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
921 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
922 h->pps.scaling_matrix4[i][x]) << shift;
923 }
924 }
925 }
926
927 static void init_dequant_tables(H264Context *h){
928 int i,x;
929 init_dequant4_coeff_table(h);
930 if(h->pps.transform_8x8_mode)
931 init_dequant8_coeff_table(h);
932 if(h->sps.transform_bypass){
933 for(i=0; i<6; i++)
934 for(x=0; x<16; x++)
935 h->dequant4_coeff[i][0][x] = 1<<6;
936 if(h->pps.transform_8x8_mode)
937 for(i=0; i<6; i++)
938 for(x=0; x<64; x++)
939 h->dequant8_coeff[i][0][x] = 1<<6;
940 }
941 }
942
943
944 int ff_h264_alloc_tables(H264Context *h){
945 MpegEncContext * const s = &h->s;
946 const int big_mb_num= s->mb_stride * (s->mb_height+1);
947 const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
948 int x,y;
949
950 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8 * sizeof(uint8_t), fail)
951
952 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count , big_mb_num * 48 * sizeof(uint8_t), fail)
953 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
954 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
955
956 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
957 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
958 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
959 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
960 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
961
962 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
963 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
964
965 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy , big_mb_num * sizeof(uint32_t), fail);
966 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
967 for(y=0; y<s->mb_height; y++){
968 for(x=0; x<s->mb_width; x++){
969 const int mb_xy= x + y*s->mb_stride;
970 const int b_xy = 4*x + 4*y*h->b_stride;
971
972 h->mb2b_xy [mb_xy]= b_xy;
973 h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
974 }
975 }
976
977 s->obmc_scratchpad = NULL;
978
979 if(!h->dequant4_coeff[0])
980 init_dequant_tables(h);
981
982 return 0;
983 fail:
984 free_tables(h, 1);
985 return -1;
986 }
987
988 /**
989 * Mimic alloc_tables(), but for every context thread.
990 */
991 static void clone_tables(H264Context *dst, H264Context *src, int i){
992 MpegEncContext * const s = &src->s;
993 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
994 dst->non_zero_count = src->non_zero_count;
995 dst->slice_table = src->slice_table;
996 dst->cbp_table = src->cbp_table;
997 dst->mb2b_xy = src->mb2b_xy;
998 dst->mb2br_xy = src->mb2br_xy;
999 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
1000 dst->mvd_table[0] = src->mvd_table[0] + i*8*2*s->mb_stride;
1001 dst->mvd_table[1] = src->mvd_table[1] + i*8*2*s->mb_stride;
1002 dst->direct_table = src->direct_table;
1003 dst->list_counts = src->list_counts;
1004
1005 dst->s.obmc_scratchpad = NULL;
1006 ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma, src->sps.chroma_format_idc);
1007 }
1008
1009 /**
1010 * Init context
1011 * Allocate buffers which are not shared amongst multiple threads.
1012 */
1013 static int context_init(H264Context *h){
1014 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1015 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1016
1017 h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
1018 h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
1019
1020 return 0;
1021 fail:
1022 return -1; // free_tables will clean up for us
1023 }
1024
1025 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
1026
1027 static av_cold void common_init(H264Context *h){
1028 MpegEncContext * const s = &h->s;
1029
1030 s->width = s->avctx->width;
1031 s->height = s->avctx->height;
1032 s->codec_id= s->avctx->codec->id;
1033
1034 ff_h264dsp_init(&h->h264dsp, 8, 1);
1035 ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
1036
1037 h->dequant_coeff_pps= -1;
1038 s->unrestricted_mv=1;
1039
1040 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
1041
1042 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
1043 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
1044 }
1045
1046 int ff_h264_decode_extradata(H264Context *h)
1047 {
1048 AVCodecContext *avctx = h->s.avctx;
1049
1050 if(avctx->extradata[0] == 1){
1051 int i, cnt, nalsize;
1052 unsigned char *p = avctx->extradata;
1053
1054 h->is_avc = 1;
1055
1056 if(avctx->extradata_size < 7) {
1057 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1058 return -1;
1059 }
1060 /* sps and pps in the avcC always have length coded with 2 bytes,
1061 so put a fake nal_length_size = 2 while parsing them */
1062 h->nal_length_size = 2;
1063 // Decode sps from avcC
1064 cnt = *(p+5) & 0x1f; // Number of sps
1065 p += 6;
1066 for (i = 0; i < cnt; i++) {
1067 nalsize = AV_RB16(p) + 2;
1068 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1069 return -1;
1070 if(decode_nal_units(h, p, nalsize) < 0) {
1071 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1072 return -1;
1073 }
1074 p += nalsize;
1075 }
1076 // Decode pps from avcC
1077 cnt = *(p++); // Number of pps
1078 for (i = 0; i < cnt; i++) {
1079 nalsize = AV_RB16(p) + 2;
1080 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1081 return -1;
1082 if (decode_nal_units(h, p, nalsize) < 0) {
1083 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1084 return -1;
1085 }
1086 p += nalsize;
1087 }
1088 // Now store right nal length size, that will be use to parse all other nals
1089 h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
1090 } else {
1091 h->is_avc = 0;
1092 if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
1093 return -1;
1094 }
1095 return 0;
1096 }
1097
1098 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1099 H264Context *h= avctx->priv_data;
1100 MpegEncContext * const s = &h->s;
1101 int i;
1102
1103 MPV_decode_defaults(s);
1104
1105 s->avctx = avctx;
1106 common_init(h);
1107
1108 s->out_format = FMT_H264;
1109 s->workaround_bugs= avctx->workaround_bugs;
1110
1111 // set defaults
1112 // s->decode_mb= ff_h263_decode_mb;
1113 s->quarter_sample = 1;
1114 if(!avctx->has_b_frames)
1115 s->low_delay= 1;
1116
1117 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1118
1119 ff_h264_decode_init_vlc();
1120
1121 h->pixel_shift = 0;
1122 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1123
1124 h->thread_context[0] = h;
1125 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1126 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1127 h->last_pocs[i] = INT_MIN;
1128 h->prev_poc_msb= 1<<16;
1129 h->x264_build = -1;
1130 ff_h264_reset_sei(h);
1131 if(avctx->codec_id == CODEC_ID_H264){
1132 if(avctx->ticks_per_frame == 1){
1133 s->avctx->time_base.den *=2;
1134 }
1135 avctx->ticks_per_frame = 2;
1136 }
1137
1138 if(avctx->extradata_size > 0 && avctx->extradata &&
1139 ff_h264_decode_extradata(h))
1140 return -1;
1141
1142 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1143 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1144 s->low_delay = 0;
1145 }
1146
1147 return 0;
1148 }
1149
1150 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1151 static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1152 {
1153 int i;
1154
1155 for (i=0; i<count; i++){
1156 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1157 IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1158 !from[i]));
1159 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1160 }
1161 }
1162
1163 static void copy_parameter_set(void **to, void **from, int count, int size)
1164 {
1165 int i;
1166
1167 for (i=0; i<count; i++){
1168 if (to[i] && !from[i]) av_freep(&to[i]);
1169 else if (from[i] && !to[i]) to[i] = av_malloc(size);
1170
1171 if (from[i]) memcpy(to[i], from[i], size);
1172 }
1173 }
1174
1175 static int decode_init_thread_copy(AVCodecContext *avctx){
1176 H264Context *h= avctx->priv_data;
1177
1178 if (!avctx->internal->is_copy)
1179 return 0;
1180 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1181 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1182
1183 return 0;
1184 }
1185
1186 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1187 static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1188 H264Context *h= dst->priv_data, *h1= src->priv_data;
1189 MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1190 int inited = s->context_initialized, err;
1191 int i;
1192
1193 if(dst == src || !s1->context_initialized) return 0;
1194
1195 err = ff_mpeg_update_thread_context(dst, src);
1196 if(err) return err;
1197
1198 //FIXME handle width/height changing
1199 if(!inited){
1200 for(i = 0; i < MAX_SPS_COUNT; i++)
1201 av_freep(h->sps_buffers + i);
1202
1203 for(i = 0; i < MAX_PPS_COUNT; i++)
1204 av_freep(h->pps_buffers + i);
1205
1206 memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1207 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1208 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1209 if (ff_h264_alloc_tables(h) < 0) {
1210 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1211 return AVERROR(ENOMEM);
1212 }
1213 context_init(h);
1214
1215 for(i=0; i<2; i++){
1216 h->rbsp_buffer[i] = NULL;
1217 h->rbsp_buffer_size[i] = 0;
1218 }
1219
1220 h->thread_context[0] = h;
1221
1222 // frame_start may not be called for the next thread (if it's decoding a bottom field)
1223 // so this has to be allocated here
1224 h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1225
1226 s->dsp.clear_blocks(h->mb);
1227 s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
1228 }
1229
1230 //extradata/NAL handling
1231 h->is_avc = h1->is_avc;
1232
1233 //SPS/PPS
1234 copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1235 h->sps = h1->sps;
1236 copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1237 h->pps = h1->pps;
1238
1239 //Dequantization matrices
1240 //FIXME these are big - can they be only copied when PPS changes?
1241 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1242
1243 for(i=0; i<6; i++)
1244 h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1245
1246 for(i=0; i<6; i++)
1247 h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1248
1249 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1250
1251 //POC timing
1252 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1253
1254 //reference lists
1255 copy_fields(h, h1, ref_count, list_count);
1256 copy_fields(h, h1, ref_list, intra_gb);
1257 copy_fields(h, h1, short_ref, cabac_init_idc);
1258
1259 copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1260 copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1261 copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1262
1263 h->last_slice_type = h1->last_slice_type;
1264
1265 if(!s->current_picture_ptr) return 0;
1266
1267 if(!s->dropable) {
1268 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1269 h->prev_poc_msb = h->poc_msb;
1270 h->prev_poc_lsb = h->poc_lsb;
1271 }
1272 h->prev_frame_num_offset= h->frame_num_offset;
1273 h->prev_frame_num = h->frame_num;
1274 h->outputed_poc = h->next_outputed_poc;
1275
1276 return err;
1277 }
1278
1279 int ff_h264_frame_start(H264Context *h){
1280 MpegEncContext * const s = &h->s;
1281 int i;
1282 const int pixel_shift = h->pixel_shift;
1283 int thread_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1;
1284
1285 if(MPV_frame_start(s, s->avctx) < 0)
1286 return -1;
1287 ff_er_frame_start(s);
1288 /*
1289 * MPV_frame_start uses pict_type to derive key_frame.
1290 * This is incorrect for H.264; IDR markings must be used.
1291 * Zero here; IDR markings per slice in frame or fields are ORed in later.
1292 * See decode_nal_units().
1293 */
1294 s->current_picture_ptr->f.key_frame = 0;
1295 s->current_picture_ptr->mmco_reset= 0;
1296
1297 assert(s->linesize && s->uvlinesize);
1298
1299 for(i=0; i<16; i++){
1300 h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
1301 h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
1302 }
1303 for(i=0; i<16; i++){
1304 h->block_offset[16+i]=
1305 h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1306 h->block_offset[48+16+i]=
1307 h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1308 }
1309
1310 /* can't be in alloc_tables because linesize isn't known there.
1311 * FIXME: redo bipred weight to not require extra buffer? */
1312 for(i = 0; i < thread_count; i++)
1313 if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1314 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1315
1316 /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
1317 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
1318
1319 // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.f.reference /*|| h->contains_intra*/ || 1;
1320
1321 // We mark the current picture as non-reference after allocating it, so
1322 // that if we break out due to an error it can be released automatically
1323 // in the next MPV_frame_start().
1324 // SVQ3 as well as most other codecs have only last/next/current and thus
1325 // get released even with set reference, besides SVQ3 and others do not
1326 // mark frames as reference later "naturally".
1327 if(s->codec_id != CODEC_ID_SVQ3)
1328 s->current_picture_ptr->f.reference = 0;
1329
1330 s->current_picture_ptr->field_poc[0]=
1331 s->current_picture_ptr->field_poc[1]= INT_MAX;
1332
1333 h->next_output_pic = NULL;
1334
1335 assert(s->current_picture_ptr->long_ref==0);
1336
1337 return 0;
1338 }
1339
1340 /**
1341 * Run setup operations that must be run after slice header decoding.
1342 * This includes finding the next displayed frame.
1343 *
1344 * @param h h264 master context
1345 * @param setup_finished enough NALs have been read that we can call
1346 * ff_thread_finish_setup()
1347 */
1348 static void decode_postinit(H264Context *h, int setup_finished){
1349 MpegEncContext * const s = &h->s;
1350 Picture *out = s->current_picture_ptr;
1351 Picture *cur = s->current_picture_ptr;
1352 int i, pics, out_of_order, out_idx;
1353 int invalid = 0, cnt = 0;
1354
1355 s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1356 s->current_picture_ptr->f.pict_type = s->pict_type;
1357
1358 if (h->next_output_pic) return;
1359
1360 if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1361 //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1362 //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1363 //to find this yet, so we assume the worst for now.
1364 //if (setup_finished)
1365 // ff_thread_finish_setup(s->avctx);
1366 return;
1367 }
1368
1369 cur->f.interlaced_frame = 0;
1370 cur->f.repeat_pict = 0;
1371
1372 /* Signal interlacing information externally. */
1373 /* Prioritize picture timing SEI information over used decoding process if it exists. */
1374
1375 if(h->sps.pic_struct_present_flag){
1376 switch (h->sei_pic_struct)
1377 {
1378 case SEI_PIC_STRUCT_FRAME:
1379 break;
1380 case SEI_PIC_STRUCT_TOP_FIELD:
1381 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1382 cur->f.interlaced_frame = 1;
1383 break;
1384 case SEI_PIC_STRUCT_TOP_BOTTOM:
1385 case SEI_PIC_STRUCT_BOTTOM_TOP:
1386 if (FIELD_OR_MBAFF_PICTURE)
1387 cur->f.interlaced_frame = 1;
1388 else
1389 // try to flag soft telecine progressive
1390 cur->f.interlaced_frame = h->prev_interlaced_frame;
1391 break;
1392 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1393 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1394 // Signal the possibility of telecined film externally (pic_struct 5,6)
1395 // From these hints, let the applications decide if they apply deinterlacing.
1396 cur->f.repeat_pict = 1;
1397 break;
1398 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1399 // Force progressive here, as doubling interlaced frame is a bad idea.
1400 cur->f.repeat_pict = 2;
1401 break;
1402 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1403 cur->f.repeat_pict = 4;
1404 break;
1405 }
1406
1407 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1408 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1409 }else{
1410 /* Derive interlacing flag from used decoding process. */
1411 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1412 }
1413 h->prev_interlaced_frame = cur->f.interlaced_frame;
1414
1415 if (cur->field_poc[0] != cur->field_poc[1]){
1416 /* Derive top_field_first from field pocs. */
1417 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1418 }else{
1419 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1420 /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1421 if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1422 || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1423 cur->f.top_field_first = 1;
1424 else
1425 cur->f.top_field_first = 0;
1426 }else{
1427 /* Most likely progressive */
1428 cur->f.top_field_first = 0;
1429 }
1430 }
1431
1432 //FIXME do something with unavailable reference frames
1433
1434 /* Sort B-frames into display order */
1435
1436 if(h->sps.bitstream_restriction_flag
1437 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1438 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1439 s->low_delay = 0;
1440 }
1441
1442 if( s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1443 && !h->sps.bitstream_restriction_flag){
1444 s->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
1445 s->low_delay= 0;
1446 }
1447
1448 pics = 0;
1449 while(h->delayed_pic[pics]) pics++;
1450
1451 assert(pics <= MAX_DELAYED_PIC_COUNT);
1452
1453 h->delayed_pic[pics++] = cur;
1454 if (cur->f.reference == 0)
1455 cur->f.reference = DELAYED_PIC_REF;
1456
1457 /* Frame reordering. This code takes pictures from coding order and sorts
1458 * them by their incremental POC value into display order. It supports POC
1459 * gaps, MMCO reset codes and random resets.
1460 * A "display group" can start either with a IDR frame (f.key_frame = 1),
1461 * and/or can be closed down with a MMCO reset code. In sequences where
1462 * there is no delay, we can't detect that (since the frame was already
1463 * output to the user), so we also set h->mmco_reset to detect the MMCO
1464 * reset code.
1465 * FIXME: if we detect insufficient delays (as per s->avctx->has_b_frames),
1466 * we increase the delay between input and output. All frames affected by
1467 * the lag (e.g. those that should have been output before another frame
1468 * that we already returned to the user) will be dropped. This is a bug
1469 * that we will fix later. */
1470 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
1471 cnt += out->poc < h->last_pocs[i];
1472 invalid += out->poc == INT_MIN;
1473 }
1474 if (!h->mmco_reset && !cur->f.key_frame && cnt + invalid == MAX_DELAYED_PIC_COUNT && cnt > 0) {
1475 h->mmco_reset = 2;
1476 if (pics > 1)
1477 h->delayed_pic[pics - 2]->mmco_reset = 2;
1478 }
1479 if (h->mmco_reset || cur->f.key_frame) {
1480 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1481 h->last_pocs[i] = INT_MIN;
1482 cnt = 0;
1483 invalid = MAX_DELAYED_PIC_COUNT;
1484 }
1485 out = h->delayed_pic[0];
1486 out_idx = 0;
1487 for (i = 1; i < MAX_DELAYED_PIC_COUNT && h->delayed_pic[i] &&
1488 !h->delayed_pic[i-1]->mmco_reset && !h->delayed_pic[i]->f.key_frame; i++)
1489 {
1490 if(h->delayed_pic[i]->poc < out->poc){
1491 out = h->delayed_pic[i];
1492 out_idx = i;
1493 }
1494 }
1495 if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->mmco_reset))
1496 h->next_outputed_poc = INT_MIN;
1497 out_of_order = !out->f.key_frame && !h->mmco_reset && (out->poc < h->next_outputed_poc);
1498
1499 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
1500 { }
1501 else if (out_of_order && pics-1 == s->avctx->has_b_frames &&
1502 s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
1503 if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
1504 s->avctx->has_b_frames = FFMAX(s->avctx->has_b_frames, cnt);
1505 }
1506 s->low_delay = 0;
1507 } else if (s->low_delay &&
1508 ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2) ||
1509 cur->f.pict_type == AV_PICTURE_TYPE_B)) {
1510 s->low_delay = 0;
1511 s->avctx->has_b_frames++;
1512 }
1513
1514 if(pics > s->avctx->has_b_frames){
1515 out->f.reference &= ~DELAYED_PIC_REF;
1516 out->owner2 = s; // for frame threading, the owner must be the second field's thread
1517 // or else the first thread can release the picture and reuse it unsafely
1518 for(i=out_idx; h->delayed_pic[i]; i++)
1519 h->delayed_pic[i] = h->delayed_pic[i+1];
1520 }
1521 memmove(h->last_pocs, &h->last_pocs[1], sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
1522 h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = cur->poc;
1523 if(!out_of_order && pics > s->avctx->has_b_frames){
1524 h->next_output_pic = out;
1525 if (out->mmco_reset) {
1526 if (out_idx > 0) {
1527 h->next_outputed_poc = out->poc;
1528 h->delayed_pic[out_idx - 1]->mmco_reset = out->mmco_reset;
1529 } else {
1530 h->next_outputed_poc = INT_MIN;
1531 }
1532 } else {
1533 if (out_idx == 0 && pics > 1 && h->delayed_pic[0]->f.key_frame) {
1534 h->next_outputed_poc = INT_MIN;
1535 } else {
1536 h->next_outputed_poc = out->poc;
1537 }
1538 }
1539 h->mmco_reset = 0;
1540 }else{
1541 av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1542 }
1543
1544 if (setup_finished)
1545 ff_thread_finish_setup(s->avctx);
1546 }
1547
1548 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1549 uint8_t *src_cb, uint8_t *src_cr,
1550 int linesize, int uvlinesize, int simple)
1551 {
1552 MpegEncContext * const s = &h->s;
1553 uint8_t *top_border;
1554 int top_idx = 1;
1555 const int pixel_shift = h->pixel_shift;
1556 int chroma444 = CHROMA444;
1557 int chroma422 = CHROMA422;
1558
1559 src_y -= linesize;
1560 src_cb -= uvlinesize;
1561 src_cr -= uvlinesize;
1562
1563 if(!simple && FRAME_MBAFF){
1564 if(s->mb_y&1){
1565 if(!MB_MBAFF){
1566 top_border = h->top_borders[0][s->mb_x];
1567 AV_COPY128(top_border, src_y + 15*linesize);
1568 if (pixel_shift)
1569 AV_COPY128(top_border+16, src_y+15*linesize+16);
1570 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1571 if(chroma444){
1572 if (pixel_shift){
1573 AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1574 AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1575 AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1576 AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1577 } else {
1578 AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1579 AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1580 }
1581 } else if(chroma422) {
1582 if (pixel_shift) {
1583 AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1584 AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
1585 } else {
1586 AV_COPY64(top_border+16, src_cb + 15*uvlinesize);
1587 AV_COPY64(top_border+24, src_cr + 15*uvlinesize);
1588 }
1589 } else {
1590 if (pixel_shift) {
1591 AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1592 AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1593 } else {
1594 AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1595 AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1596 }
1597 }
1598 }
1599 }
1600 }else if(MB_MBAFF){
1601 top_idx = 0;
1602 }else
1603 return;
1604 }
1605
1606 top_border = h->top_borders[top_idx][s->mb_x];
1607 // There are two lines saved, the line above the the top macroblock of a pair,
1608 // and the line above the bottom macroblock
1609 AV_COPY128(top_border, src_y + 16*linesize);
1610 if (pixel_shift)
1611 AV_COPY128(top_border+16, src_y+16*linesize+16);
1612
1613 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1614 if(chroma444){
1615 if (pixel_shift){
1616 AV_COPY128(top_border+32, src_cb + 16*linesize);
1617 AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1618 AV_COPY128(top_border+64, src_cr + 16*linesize);
1619 AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1620 } else {
1621 AV_COPY128(top_border+16, src_cb + 16*linesize);
1622 AV_COPY128(top_border+32, src_cr + 16*linesize);
1623 }
1624 } else if(chroma422) {
1625 if (pixel_shift) {
1626 AV_COPY128(top_border+32, src_cb+16*uvlinesize);
1627 AV_COPY128(top_border+48, src_cr+16*uvlinesize);
1628 } else {
1629 AV_COPY64(top_border+16, src_cb+16*uvlinesize);
1630 AV_COPY64(top_border+24, src_cr+16*uvlinesize);
1631 }
1632 } else {
1633 if (pixel_shift) {
1634 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1635 AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1636 } else {
1637 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1638 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1639 }
1640 }
1641 }
1642 }
1643
1644 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1645 uint8_t *src_cb, uint8_t *src_cr,
1646 int linesize, int uvlinesize,
1647 int xchg, int chroma444,
1648 int simple, int pixel_shift){
1649 MpegEncContext * const s = &h->s;
1650 int deblock_topleft;
1651 int deblock_top;
1652 int top_idx = 1;
1653 uint8_t *top_border_m1;
1654 uint8_t *top_border;
1655
1656 if(!simple && FRAME_MBAFF){
1657 if(s->mb_y&1){
1658 if(!MB_MBAFF)
1659 return;
1660 }else{
1661 top_idx = MB_MBAFF ? 0 : 1;
1662 }
1663 }
1664
1665 if(h->deblocking_filter == 2) {
1666 deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1667 deblock_top = h->top_type;
1668 } else {
1669 deblock_topleft = (s->mb_x > 0);
1670 deblock_top = (s->mb_y > !!MB_FIELD);
1671 }
1672
1673 src_y -= linesize + 1 + pixel_shift;
1674 src_cb -= uvlinesize + 1 + pixel_shift;
1675 src_cr -= uvlinesize + 1 + pixel_shift;
1676
1677 top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1678 top_border = h->top_borders[top_idx][s->mb_x];
1679
1680 #define XCHG(a,b,xchg)\
1681 if (pixel_shift) {\
1682 if (xchg) {\
1683 AV_SWAP64(b+0,a+0);\
1684 AV_SWAP64(b+8,a+8);\
1685 } else {\
1686 AV_COPY128(b,a); \
1687 }\
1688 } else \
1689 if (xchg) AV_SWAP64(b,a);\
1690 else AV_COPY64(b,a);
1691
1692 if(deblock_top){
1693 if(deblock_topleft){
1694 XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
1695 }
1696 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1697 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1698 if(s->mb_x+1 < s->mb_width){
1699 XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
1700 }
1701 }
1702 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1703 if(chroma444){
1704 if(deblock_topleft){
1705 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1706 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1707 }
1708 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1709 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1710 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1711 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1712 if(s->mb_x+1 < s->mb_width){
1713 XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1714 XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1715 }
1716 } else {
1717 if(deblock_top){
1718 if(deblock_topleft){
1719 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1720 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1721 }
1722 XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1723 XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
1724 }
1725 }
1726 }
1727 }
1728
1729 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1730 if (high_bit_depth) {
1731 return AV_RN32A(((int32_t*)mb) + index);
1732 } else
1733 return AV_RN16A(mb + index);
1734 }
1735
1736 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1737 if (high_bit_depth) {
1738 AV_WN32A(((int32_t*)mb) + index, value);
1739 } else
1740 AV_WN16A(mb + index, value);
1741 }
1742
1743 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1744 int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1745 {
1746 MpegEncContext * const s = &h->s;
1747 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1748 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1749 int i;
1750 int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1751 block_offset += 16*p;
1752 if(IS_INTRA4x4(mb_type)){
1753 if(simple || !s->encoding){
1754 if(IS_8x8DCT(mb_type)){
1755 if(transform_bypass){
1756 idct_dc_add =
1757 idct_add = s->dsp.add_pixels8;
1758 }else{
1759 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1760 idct_add = h->h264dsp.h264_idct8_add;
1761 }
1762 for(i=0; i<16; i+=4){
1763 uint8_t * const ptr= dest_y + block_offset[i];
1764 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1765 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1766 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1767 }else{
1768 const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1769 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1770 (h->topright_samples_available<<i)&0x4000, linesize);
1771 if(nnz){
1772 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1773 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1774 else
1775 idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1776 }
1777 }
1778 }
1779 }else{
1780 if(transform_bypass){
1781 idct_dc_add =
1782 idct_add = s->dsp.add_pixels4;
1783 }else{
1784 idct_dc_add = h->h264dsp.h264_idct_dc_add;
1785 idct_add = h->h264dsp.h264_idct_add;
1786 }
1787 for(i=0; i<16; i++){
1788 uint8_t * const ptr= dest_y + block_offset[i];
1789 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1790
1791 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1792 h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1793 }else{
1794 uint8_t *topright;
1795 int nnz, tr;
1796 uint64_t tr_high;
1797 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1798 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1799 assert(s->mb_y || linesize <= block_offset[i]);
1800 if(!topright_avail){
1801 if (pixel_shift) {
1802 tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1803 topright= (uint8_t*) &tr_high;
1804 } else {
1805 tr= ptr[3 - linesize]*0x01010101u;
1806 topright= (uint8_t*) &tr;
1807 }
1808 }else
1809 topright= ptr + (4 << pixel_shift) - linesize;
1810 }else
1811 topright= NULL;
1812
1813 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1814 nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1815 if(nnz){
1816 if(is_h264){
1817 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1818 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1819 else
1820 idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1821 }else
1822 ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1823 }
1824 }
1825 }
1826 }
1827 }
1828 }else{
1829 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1830 if(is_h264){
1831 if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
1832 if(!transform_bypass)
1833 h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1834 else{
1835 static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1836 8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1837 for(i = 0; i < 16; i++)
1838 dctcoef_set(h->mb+(p*256 << pixel_shift), pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
1839 }
1840 }
1841 }else
1842 ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1843 }
1844 }
1845
1846 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1847 int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1848 {
1849 MpegEncContext * const s = &h->s;
1850 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1851 int i;
1852 block_offset += 16*p;
1853 if(!IS_INTRA4x4(mb_type)){
1854 if(is_h264){
1855 if(IS_INTRA16x16(mb_type)){
1856 if(transform_bypass){
1857 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1858 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1859 }else{
1860 for(i=0; i<16; i++){
1861 if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1862 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1863 }
1864 }
1865 }else{
1866 h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1867 }
1868 }else if(h->cbp&15){
1869 if(transform_bypass){
1870 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1871 idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1872 for(i=0; i<16; i+=di){
1873 if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1874 idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1875 }
1876 }
1877 }else{
1878 if(IS_8x8DCT(mb_type)){
1879 h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1880 }else{
1881 h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1882 }
1883 }
1884 }
1885 }else{
1886 for(i=0; i<16; i++){
1887 if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1888 uint8_t * const ptr= dest_y + block_offset[i];
1889 ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1890 }
1891 }
1892 }
1893 }
1894 }
1895
1896 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift)
1897 {
1898 MpegEncContext * const s = &h->s;
1899 const int mb_x= s->mb_x;
1900 const int mb_y= s->mb_y;
1901 const int mb_xy= h->mb_xy;
1902 const int mb_type = s->current_picture.f.mb_type[mb_xy];
1903 uint8_t *dest_y, *dest_cb, *dest_cr;
1904 int linesize, uvlinesize /*dct_offset*/;
1905 int i, j;
1906 int *block_offset = &h->block_offset[0];
1907 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1908 /* is_h264 should always be true if SVQ3 is disabled. */
1909 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1910 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1911 const int block_h = 16 >> s->chroma_y_shift;
1912 const int chroma422 = CHROMA422;
1913
1914 dest_y = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
1915 dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1916 dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1917
1918 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1919 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
1920
1921 h->list_counts[mb_xy]= h->list_count;
1922
1923 if (!simple && MB_FIELD) {
1924 linesize = h->mb_linesize = s->linesize * 2;
1925 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1926 block_offset = &h->block_offset[48];
1927 if(mb_y&1){ //FIXME move out of this function?
1928 dest_y -= s->linesize*15;
1929 dest_cb-= s->uvlinesize * (block_h - 1);
1930 dest_cr-= s->uvlinesize * (block_h - 1);
1931 }
1932 if(FRAME_MBAFF) {
1933 int list;
1934 for(list=0; list<h->list_count; list++){
1935 if(!USES_LIST(mb_type, list))
1936 continue;
1937 if(IS_16X16(mb_type)){
1938 int8_t *ref = &h->ref_cache[list][scan8[0]];
1939 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1940 }else{
1941 for(i=0; i<16; i+=4){
1942 int ref = h->ref_cache[list][scan8[i]];
1943 if(ref >= 0)
1944 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1945 }
1946 }
1947 }
1948 }
1949 } else {
1950 linesize = h->mb_linesize = s->linesize;
1951 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1952 // dct_offset = s->linesize * 16;
1953 }
1954
1955 if (!simple && IS_INTRA_PCM(mb_type)) {
1956 if (pixel_shift) {
1957 const int bit_depth = h->sps.bit_depth_luma;
1958 int j;
1959 GetBitContext gb;
1960 init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1961
1962 for (i = 0; i < 16; i++) {
1963 uint16_t *tmp_y = (uint16_t*)(dest_y + i*linesize);
1964 for (j = 0; j < 16; j++)
1965 tmp_y[j] = get_bits(&gb, bit_depth);
1966 }
1967 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1968 if (!h->sps.chroma_format_idc) {
1969 for (i = 0; i < block_h; i++) {
1970 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1971 for (j = 0; j < 8; j++) {
1972 tmp_cb[j] = 1 << (bit_depth - 1);
1973 }
1974 }
1975 for (i = 0; i < block_h; i++) {
1976 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1977 for (j = 0; j < 8; j++) {
1978 tmp_cr[j] = 1 << (bit_depth - 1);
1979 }
1980 }
1981 } else {
1982 for (i = 0; i < block_h; i++) {
1983 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1984 for (j = 0; j < 8; j++)
1985 tmp_cb[j] = get_bits(&gb, bit_depth);
1986 }
1987 for (i = 0; i < block_h; i++) {
1988 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1989 for (j = 0; j < 8; j++)
1990 tmp_cr[j] = get_bits(&gb, bit_depth);
1991 }
1992 }
1993 }
1994 } else {
1995 for (i=0; i<16; i++) {
1996 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
1997 }
1998 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1999 if (!h->sps.chroma_format_idc) {
2000 for (i = 0; i < block_h; i++) {
2001 memset(dest_cb + i*uvlinesize, 128, 8);
2002 memset(dest_cr + i*uvlinesize, 128, 8);
2003 }
2004 } else {
2005 for (i = 0; i < block_h; i++) {
2006 memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4, 8);
2007 memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4, 8);
2008 }
2009 }
2010 }
2011 }
2012 } else {
2013 if(IS_INTRA(mb_type)){
2014 if(h->deblocking_filter)
2015 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
2016
2017 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2018 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2019 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2020 }
2021
2022 hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
2023
2024 if(h->deblocking_filter)
2025 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
2026 }else if(is_h264){
2027 if (chroma422) {
2028 hl_motion_422(h, dest_y, dest_cb, dest_cr,
2029 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2030 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2031 h->h264dsp.weight_h264_pixels_tab,
2032 h->h264dsp.biweight_h264_pixels_tab,
2033 pixel_shift);
2034 } else {
2035 hl_motion_420(h, dest_y, dest_cb, dest_cr,
2036 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2037 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2038 h->h264dsp.weight_h264_pixels_tab,
2039 h->h264dsp.biweight_h264_pixels_tab,
2040 pixel_shift);
2041 }
2042 }
2043
2044 hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
2045
2046 if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2047 uint8_t *dest[2] = {dest_cb, dest_cr};
2048 if(transform_bypass){
2049 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2050 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
2051 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
2052 }else{
2053 idct_add = s->dsp.add_pixels4;
2054 for(j=1; j<3; j++){
2055 for(i=j*16; i<j*16+4; i++){
2056 if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2057 idct_add (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
2058 }
2059 if (chroma422) {
2060 for(i=j*16+4; i<j*16+8; i++){
2061 if(h->non_zero_count_cache[ scan8[i+4] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2062 idct_add (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
2063 }
2064 }
2065 }
2066 }
2067 }else{
2068 if(is_h264){
2069 int qp[2];
2070 if (chroma422) {
2071 qp[0] = h->chroma_qp[0] + 3;
2072 qp[1] = h->chroma_qp[1] + 3;
2073 } else {
2074 qp[0] = h->chroma_qp[0];
2075 qp[1] = h->chroma_qp[1];
2076 }
2077 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
2078 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]);
2079 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
2080 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]);
2081 h->h264dsp.h264_idct_add8(dest, block_offset,
2082 h->mb, uvlinesize,
2083 h->non_zero_count_cache);
2084 }else{
2085 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]);
2086 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]);
2087 for(j=1; j<3; j++){
2088 for(i=j*16; i<j*16+4; i++){
2089 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2090 uint8_t * const ptr= dest[j-1] + block_offset[i];
2091 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
2092 }
2093 }
2094 }
2095 }
2096 }
2097 }
2098 }
2099 if(h->cbp || IS_INTRA(mb_type))
2100 {
2101 s->dsp.clear_blocks(h->mb);
2102 s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2103 }
2104 }
2105
2106 static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
2107 MpegEncContext * const s = &h->s;
2108 const int mb_x= s->mb_x;
2109 const int mb_y= s->mb_y;
2110 const int mb_xy= h->mb_xy;
2111 const int mb_type = s->current_picture.f.mb_type[mb_xy];
2112 uint8_t *dest[3];
2113 int linesize;
2114 int i, j, p;
2115 int *block_offset = &h->block_offset[0];
2116 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2117 const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
2118
2119 for (p = 0; p < plane_count; p++)
2120 {
2121 dest[p] = s->current_picture.f.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
2122 s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
2123 }
2124
2125 h->list_counts[mb_xy]= h->list_count;
2126
2127 if (!simple && MB_FIELD) {
2128 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
2129 block_offset = &h->block_offset[48];
2130 if(mb_y&1) //FIXME move out of this function?
2131 for (p = 0; p < 3; p++)
2132 dest[p] -= s->linesize*15;
2133 if(FRAME_MBAFF) {
2134 int list;
2135 for(list=0; list<h->list_count; list++){
2136 if(!USES_LIST(mb_type, list))
2137 continue;
2138 if(IS_16X16(mb_type)){
2139 int8_t *ref = &h->ref_cache[list][scan8[0]];
2140 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2141 }else{
2142 for(i=0; i<16; i+=4){
2143 int ref = h->ref_cache[list][scan8[i]];
2144 if(ref >= 0)
2145 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2146 }
2147 }
2148 }
2149 }
2150 } else {
2151 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2152 }
2153
2154 if (!simple && IS_INTRA_PCM(mb_type)) {
2155 if (pixel_shift) {
2156 const int bit_depth = h->sps.bit_depth_luma;
2157 GetBitContext gb;
2158 init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2159
2160 for (p = 0; p < plane_count; p++) {
2161 for (i = 0; i < 16; i++) {
2162 uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2163 for (j = 0; j < 16; j++)
2164 tmp[j] = get_bits(&gb, bit_depth);
2165 }
2166 }
2167 } else {
2168 for (p = 0; p < plane_count; p++) {
2169 for (i = 0; i < 16; i++) {
2170 memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2171 }
2172 }
2173 }
2174 } else {
2175 if(IS_INTRA(mb_type)){
2176 if(h->deblocking_filter)
2177 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2178
2179 for (p = 0; p < plane_count; p++)
2180 hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2181
2182 if(h->deblocking_filter)
2183 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2184 }else{
2185 hl_motion(h, dest[0], dest[1], dest[2],
2186 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2187 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2188 h->h264dsp.weight_h264_pixels_tab,
2189 h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 3);
2190 }
2191
2192 for (p = 0; p < plane_count; p++)
2193 hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2194 }
2195 if(h->cbp || IS_INTRA(mb_type))
2196 {
2197 s->dsp.clear_blocks(h->mb);
2198 s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2199 }
2200 }
2201
2202 /**
2203 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2204 */
2205 #define hl_decode_mb_simple(sh, bits) \
2206 static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2207 hl_decode_mb_internal(h, 1, sh); \
2208 }
2209 hl_decode_mb_simple(0, 8)
2210 hl_decode_mb_simple(1, 16)
2211
2212 /**
2213 * Process a macroblock; this handles edge cases, such as interlacing.
2214 */
2215 static void av_noinline hl_decode_mb_complex(H264Context *h){
2216 hl_decode_mb_internal(h, 0, h->pixel_shift);
2217 }
2218
2219 static void av_noinline hl_decode_mb_444_complex(H264Context *h){
2220 hl_decode_mb_444_internal(h, 0, h->pixel_shift);
2221 }
2222
2223 static void av_noinline hl_decode_mb_444_simple(H264Context *h){
2224 hl_decode_mb_444_internal(h, 1, 0);
2225 }
2226
2227 void ff_h264_hl_decode_mb(H264Context *h){
2228 MpegEncContext * const s = &h->s;
2229 const int mb_xy= h->mb_xy;
2230 const int mb_type = s->current_picture.f.mb_type[mb_xy];
2231 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2232
2233 if (CHROMA444) {
2234 if(is_complex || h->pixel_shift)
2235 hl_decode_mb_444_complex(h);
2236 else
2237 hl_decode_mb_444_simple(h);
2238 } else if (is_complex) {
2239 hl_decode_mb_complex(h);
2240 } else if (h->pixel_shift) {
2241 hl_decode_mb_simple_16(h);
2242 } else
2243 hl_decode_mb_simple_8(h);
2244 }
2245
2246 static int pred_weight_table(H264Context *h){
2247 MpegEncContext * const s = &h->s;
2248 int list, i;
2249 int luma_def, chroma_def;
2250
2251 h->use_weight= 0;
2252 h->use_weight_chroma= 0;
2253 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2254 if(h->sps.chroma_format_idc)
2255 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2256 luma_def = 1<<h->luma_log2_weight_denom;
2257 chroma_def = 1<<h->chroma_log2_weight_denom;
2258
2259 for(list=0; list<2; list++){
2260 h->luma_weight_flag[list] = 0;
2261 h->chroma_weight_flag[list] = 0;
2262 for(i=0; i<h->ref_count[list]; i++){
2263 int luma_weight_flag, chroma_weight_flag;
2264
2265 luma_weight_flag= get_bits1(&s->gb);
2266 if(luma_weight_flag){
2267 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2268 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2269 if( h->luma_weight[i][list][0] != luma_def
2270 || h->luma_weight[i][list][1] != 0) {
2271 h->use_weight= 1;
2272 h->luma_weight_flag[list]= 1;
2273 }
2274 }else{
2275 h->luma_weight[i][list][0]= luma_def;
2276 h->luma_weight[i][list][1]= 0;
2277 }
2278
2279 if(h->sps.chroma_format_idc){
2280 chroma_weight_flag= get_bits1(&s->gb);
2281 if(chroma_weight_flag){
2282 int j;
2283 for(j=0; j<2; j++){
2284 h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2285 h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2286 if( h->chroma_weight[i][list][j][0] != chroma_def
2287 || h->chroma_weight[i][list][j][1] != 0) {
2288 h->use_weight_chroma= 1;
2289 h->chroma_weight_flag[list]= 1;
2290 }
2291 }
2292 }else{
2293 int j;
2294 for(j=0; j<2; j++){
2295 h->chroma_weight[i][list][j][0]= chroma_def;
2296 h->chroma_weight[i][list][j][1]= 0;
2297 }
2298 }
2299 }
2300 }
2301 if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
2302 }
2303 h->use_weight= h->use_weight || h->use_weight_chroma;
2304 return 0;
2305 }
2306
2307 /**
2308 * Initialize implicit_weight table.
2309 * @param field 0/1 initialize the weight for interlaced MBAFF
2310 * -1 initializes the rest
2311 */
2312 static void implicit_weight_table(H264Context *h, int field){
2313 MpegEncContext * const s = &h->s;
2314 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2315
2316 for (i = 0; i < 2; i++) {
2317 h->luma_weight_flag[i] = 0;
2318 h->chroma_weight_flag[i] = 0;
2319 }
2320
2321 if(field < 0){
2322 if (s->picture_structure == PICT_FRAME) {
2323 cur_poc = s->current_picture_ptr->poc;
2324 } else {
2325 cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2326 }
2327 if( h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
2328 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2329 h->use_weight= 0;
2330 h->use_weight_chroma= 0;
2331 return;
2332 }
2333 ref_start= 0;
2334 ref_count0= h->ref_count[0];
2335 ref_count1= h->ref_count[1];
2336 }else{
2337 cur_poc = s->current_picture_ptr->field_poc[field];
2338 ref_start= 16;
2339 ref_count0= 16+2*h->ref_count[0];
2340 ref_count1= 16+2*h->ref_count[1];
2341 }
2342
2343 h->use_weight= 2;
2344 h->use_weight_chroma= 2;
2345 h->luma_log2_weight_denom= 5;
2346 h->chroma_log2_weight_denom= 5;
2347
2348 for(ref0=ref_start; ref0 < ref_count0; ref0++){
2349 int poc0 = h->ref_list[0][ref0].poc;
2350 for(ref1=ref_start; ref1 < ref_count1; ref1++){
2351 int w = 32;
2352 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2353 int poc1 = h->ref_list[1][ref1].poc;
2354 int td = av_clip(poc1 - poc0, -128, 127);
2355 if(td){
2356 int tb = av_clip(cur_poc - poc0, -128, 127);
2357 int tx = (16384 + (FFABS(td) >> 1)) / td;
2358 int dist_scale_factor = (tb*tx + 32) >> 8;
2359 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2360 w = 64 - dist_scale_factor;
2361 }
2362 }
2363 if(field<0){
2364 h->implicit_weight[ref0][ref1][0]=
2365 h->implicit_weight[ref0][ref1][1]= w;
2366 }else{
2367 h->implicit_weight[ref0][ref1][field]=w;
2368 }
2369 }
2370 }
2371 }
2372
2373 /**
2374 * instantaneous decoder refresh.
2375 */
2376 static void idr(H264Context *h){
2377 ff_h264_remove_all_refs(h);
2378 h->prev_frame_num= 0;
2379 h->prev_frame_num_offset= 0;
2380 h->prev_poc_msb=
2381 h->prev_poc_lsb= 0;
2382 }
2383
2384 /* forget old pics after a seek */
2385 static void flush_dpb(AVCodecContext *avctx){
2386 H264Context *h= avctx->priv_data;
2387 int i;
2388 for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
2389 if(h->delayed_pic[i])
2390 h->delayed_pic[i]->f.reference = 0;
2391 h->delayed_pic[i]= NULL;
2392 }
2393 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2394 h->last_pocs[i] = INT_MIN;
2395 h->outputed_poc=h->next_outputed_poc= INT_MIN;
2396 h->prev_interlaced_frame = 1;
2397 idr(h);
2398 if(h->s.current_picture_ptr)
2399 h->s.current_picture_ptr->f.reference = 0;
2400 h->s.first_field= 0;
2401 ff_h264_reset_sei(h);
2402 ff_mpeg_flush(avctx);
2403 }
2404
2405 static int init_poc(H264Context *h){
2406 MpegEncContext * const s = &h->s;
2407 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2408 int field_poc[2];
2409 Picture *cur = s->current_picture_ptr;
2410
2411 h->frame_num_offset= h->prev_frame_num_offset;
2412 if(h->frame_num < h->prev_frame_num)
2413 h->frame_num_offset += max_frame_num;
2414
2415 if(h->sps.poc_type==0){
2416 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2417
2418 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2419 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2420 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2421 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2422 else
2423 h->poc_msb = h->prev_poc_msb;
2424 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2425 field_poc[0] =
2426 field_poc[1] = h->poc_msb + h->poc_lsb;
2427 if(s->picture_structure == PICT_FRAME)
2428 field_poc[1] += h->delta_poc_bottom;
2429 }else if(h->sps.poc_type==1){
2430 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2431 int i;
2432
2433 if(h->sps.poc_cycle_length != 0)
2434 abs_frame_num = h->frame_num_offset + h->frame_num;
2435 else
2436 abs_frame_num = 0;
2437
2438 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2439 abs_frame_num--;
2440
2441 expected_delta_per_poc_cycle = 0;
2442 for(i=0; i < h->sps.poc_cycle_length; i++)
2443 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2444
2445 if(abs_frame_num > 0){
2446 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2447 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2448
2449 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2450 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2451 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2452 } else
2453 expectedpoc = 0;
2454
2455 if(h->nal_ref_idc == 0)
2456 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2457
2458 field_poc[0] = expectedpoc + h->delta_poc[0];
2459 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2460
2461 if(s->picture_structure == PICT_FRAME)
2462 field_poc[1] += h->delta_poc[1];
2463 }else{
2464 int poc= 2*(h->frame_num_offset + h->frame_num);
2465
2466 if(!h->nal_ref_idc)
2467 poc--;
2468
2469 field_poc[0]= poc;
2470 field_poc[1]= poc;
2471 }
2472
2473 if(s->picture_structure != PICT_BOTTOM_FIELD)
2474 s->current_picture_ptr->field_poc[0]= field_poc[0];
2475 if(s->picture_structure != PICT_TOP_FIELD)
2476 s->current_picture_ptr->field_poc[1]= field_poc[1];
2477 cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
2478
2479 return 0;
2480 }
2481
2482
2483 /**
2484 * initialize scan tables
2485 */
2486 static void init_scan_tables(H264Context *h){
2487 int i;
2488 for(i=0; i<16; i++){
2489 #define T(x) (x>>2) | ((x<<2) & 0xF)
2490 h->zigzag_scan[i] = T(zigzag_scan[i]);
2491 h-> field_scan[i] = T( field_scan[i]);
2492 #undef T
2493 }
2494 for(i=0; i<64; i++){
2495 #define T(x) (x>>3) | ((x&7)<<3)
2496 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2497 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2498 h->field_scan8x8[i] = T(field_scan8x8[i]);
2499 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
2500 #undef T
2501 }
2502 if(h->sps.transform_bypass){ //FIXME same ugly
2503 h->zigzag_scan_q0 = zigzag_scan;
2504 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
2505 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2506 h->field_scan_q0 = field_scan;
2507 h->field_scan8x8_q0 = field_scan8x8;
2508 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
2509 }else{
2510 h->zigzag_scan_q0 = h->zigzag_scan;
2511 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
2512 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2513 h->field_scan_q0 = h->field_scan;
2514 h->field_scan8x8_q0 = h->field_scan8x8;
2515 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
2516 }
2517 }
2518
2519 static int field_end(H264Context *h, int in_setup){
2520 MpegEncContext * const s = &h->s;
2521 AVCodecContext * const avctx= s->avctx;
2522 int err = 0;
2523 s->mb_y= 0;
2524
2525 if (!in_setup && !s->dropable)
2526 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
2527 s->picture_structure==PICT_BOTTOM_FIELD);
2528
2529 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2530 ff_vdpau_h264_set_reference_frames(s);
2531
2532 if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2533 if(!s->dropable) {
2534 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2535 h->prev_poc_msb= h->poc_msb;
2536 h->prev_poc_lsb= h->poc_lsb;
2537 }
2538 h->prev_frame_num_offset= h->frame_num_offset;
2539 h->prev_frame_num= h->frame_num;
2540 h->outputed_poc = h->next_outputed_poc;
2541 }
2542
2543 if (avctx->hwaccel) {
2544 if (avctx->hwaccel->end_frame(avctx) < 0)
2545 av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2546 }
2547
2548 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2549 ff_vdpau_h264_picture_complete(s);
2550
2551 /*
2552 * FIXME: Error handling code does not seem to support interlaced
2553 * when slices span multiple rows
2554 * The ff_er_add_slice calls don't work right for bottom
2555 * fields; they cause massive erroneous error concealing
2556 * Error marking covers both fields (top and bottom).
2557 * This causes a mismatched s->error_count
2558 * and a bad error table. Further, the error count goes to
2559 * INT_MAX when called for bottom field, because mb_y is
2560 * past end by one (callers fault) and resync_mb_y != 0
2561 * causes problems for the first MB line, too.
2562 */
2563 if (!FIELD_PICTURE)
2564 ff_er_frame_end(s);
2565
2566 MPV_frame_end(s);
2567
2568 h->current_slice=0;
2569
2570 return err;
2571 }
2572
2573 /**
2574 * Replicate H264 "master" context to thread contexts.
2575 */
2576 static void clone_slice(H264Context *dst, H264Context *src)
2577 {
2578 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2579 dst->s.current_picture_ptr = src->s.current_picture_ptr;
2580 dst->s.current_picture = src->s.current_picture;
2581 dst->s.linesize = src->s.linesize;
2582 dst->s.uvlinesize = src->s.uvlinesize;
2583 dst->s.first_field = src->s.first_field;
2584
2585 dst->prev_poc_msb = src->prev_poc_msb;
2586 dst->prev_poc_lsb = src->prev_poc_lsb;
2587 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2588 dst->prev_frame_num = src->prev_frame_num;
2589 dst->short_ref_count = src->short_ref_count;
2590
2591 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2592 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2593 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2594 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
2595
2596 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2597 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
2598 }
2599
2600 /**
2601 * Compute profile from profile_idc and constraint_set?_flags.
2602 *
2603 * @param sps SPS
2604 *
2605 * @return profile as defined by FF_PROFILE_H264_*
2606 */
2607 int ff_h264_get_profile(SPS *sps)
2608 {
2609 int profile = sps->profile_idc;
2610
2611 switch(sps->profile_idc) {
2612 case FF_PROFILE_H264_BASELINE:
2613 // constraint_set1_flag set to 1
2614 profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2615 break;
2616 case FF_PROFILE_H264_HIGH_10:
2617 case FF_PROFILE_H264_HIGH_422:
2618 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2619 // constraint_set3_flag set to 1
2620 profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2621 break;
2622 }
2623
2624 return profile;
2625 }
2626
2627 /**
2628 * Decode a slice header.
2629 * This will also call MPV_common_init() and frame_start() as needed.
2630 *
2631 * @param h h264context
2632 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2633 *
2634 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2635 */
2636 static int decode_slice_header(H264Context *h, H264Context *h0){
2637 MpegEncContext * const s = &h->s;
2638 MpegEncContext * const s0 = &h0->s;
2639 unsigned int first_mb_in_slice;
2640 unsigned int pps_id;
2641 int num_ref_idx_active_override_flag;
2642 unsigned int slice_type, tmp, i, j;
2643 int default_ref_list_done = 0;
2644 int last_pic_structure;
2645
2646 s->dropable= h->nal_ref_idc == 0;
2647
2648 /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2649 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
2650 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2651 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2652 }else{
2653 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2654 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2655 }
2656
2657 first_mb_in_slice= get_ue_golomb(&s->gb);
2658
2659 if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2660 if(h0->current_slice && FIELD_PICTURE){
2661 field_end(h, 1);
2662 }
2663
2664 h0->current_slice = 0;
2665 if (!s0->first_field)
2666 s->current_picture_ptr= NULL;
2667 }
2668
2669 slice_type= get_ue_golomb_31(&s->gb);
2670 if(slice_type > 9){
2671 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);
2672 return -1;
2673 }
2674 if(slice_type > 4){
2675 slice_type -= 5;
2676 h->slice_type_fixed=1;
2677 }else
2678 h->slice_type_fixed=0;
2679
2680 slice_type= golomb_to_pict_type[ slice_type ];
2681 if (slice_type == AV_PICTURE_TYPE_I
2682 || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
2683 default_ref_list_done = 1;
2684 }
2685 h->slice_type= slice_type;
2686 h->slice_type_nos= slice_type & 3;
2687
2688 s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
2689
2690 pps_id= get_ue_golomb(&s->gb);
2691 if(pps_id>=MAX_PPS_COUNT){
2692 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2693 return -1;
2694 }
2695 if(!h0->pps_buffers[pps_id]) {
2696 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
2697 return -1;
2698 }
2699 h->pps= *h0->pps_buffers[pps_id];
2700
2701 if(!h0->sps_buffers[h->pps.sps_id]) {
2702 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
2703 return -1;
2704 }
2705 h->sps = *h0->sps_buffers[h->pps.sps_id];
2706
2707 s->avctx->profile = ff_h264_get_profile(&h->sps);
2708 s->avctx->level = h->sps.level_idc;
2709 s->avctx->refs = h->sps.ref_frame_count;
2710
2711 if(h == h0 && h->dequant_coeff_pps != pps_id){
2712 h->dequant_coeff_pps = pps_id;
2713 init_dequant_tables(h);
2714 }
2715
2716 s->mb_width= h->sps.mb_width;
2717 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2718
2719 h->b_stride= s->mb_width*4;
2720
2721 s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2722
2723 s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2724 if(h->sps.frame_mbs_only_flag)
2725 s->height= 16*s->mb_height - (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2726 else
2727 s->height= 16*s->mb_height - (2<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2728
2729 if (s->context_initialized
2730 && ( s->width != s->avctx->width || s->height != s->avctx->height
2731 || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
2732 if(h != h0) {
2733 av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
2734 return -1; // width / height changed during parallelized decoding
2735 }
2736 free_tables(h, 0);
2737 flush_dpb(s->avctx);
2738 MPV_common_end(s);
2739 }
2740 if (!s->context_initialized) {
2741 if (h != h0) {
2742 av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2743 return -1;
2744 }
2745
2746 avcodec_set_dimensions(s->avctx, s->width, s->height);
2747 s->avctx->sample_aspect_ratio= h->sps.sar;
2748 av_assert0(s->avctx->sample_aspect_ratio.den);
2749
2750 if(h->sps.video_signal_type_present_flag){
2751 s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2752 if(h->sps.colour_description_present_flag){
2753 s->avctx->color_primaries = h->sps.color_primaries;
2754 s->avctx->color_trc = h->sps.color_trc;
2755 s->avctx->colorspace = h->sps.colorspace;
2756 }
2757 }
2758
2759 if(h->sps.timing_info_present_flag){
2760 int64_t den= h->sps.time_scale;
2761 if(h->x264_build < 44U)
2762 den *= 2;
2763 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2764 h->sps.num_units_in_tick, den, 1<<30);
2765 }
2766
2767 switch (h->sps.bit_depth_luma) {
2768 case 9 :
2769 if (CHROMA444) {
2770 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2771 s->avctx->pix_fmt = PIX_FMT_GBRP9;
2772 } else
2773 s->avctx->pix_fmt = PIX_FMT_YUV444P9;
2774 } else if (CHROMA422)
2775 s->avctx->pix_fmt = PIX_FMT_YUV422P9;
2776 else
2777 s->avctx->pix_fmt = PIX_FMT_YUV420P9;
2778 break;
2779 case 10 :
2780 if (CHROMA444) {
2781 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2782 s->avctx->pix_fmt = PIX_FMT_GBRP10;
2783 } else
2784 s->avctx->pix_fmt = PIX_FMT_YUV444P10;
2785 } else if (CHROMA422)
2786 s->avctx->pix_fmt = PIX_FMT_YUV422P10;
2787 else
2788 s->avctx->pix_fmt = PIX_FMT_YUV420P10;
2789 break;
2790 default:
2791 if (CHROMA444){
2792 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2793 s->avctx->pix_fmt = PIX_FMT_GBRP;
2794 } else
2795 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
2796 } else if (CHROMA422) {
2797 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P : PIX_FMT_YUV422P;
2798 }else{
2799 s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2800 s->avctx->codec->pix_fmts ?
2801 s->avctx->codec->pix_fmts :
2802 s->avctx->color_range == AVCOL_RANGE_JPEG ?
2803 hwaccel_pixfmt_list_h264_jpeg_420 :
2804 ff_hwaccel_pixfmt_list_420);
2805 }
2806 }
2807
2808 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2809
2810 if (MPV_common_init(s) < 0) {
2811 av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed.\n");
2812 return -1;
2813 }
2814 s->first_field = 0;
2815 h->prev_interlaced_frame = 1;
2816
2817 init_scan_tables(h);
2818 if (ff_h264_alloc_tables(h) < 0) {
2819 av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2820 return AVERROR(ENOMEM);
2821 }
2822
2823 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
2824 if (context_init(h) < 0) {
2825 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2826 return -1;
2827 }
2828 } else {
2829 for(i = 1; i < s->avctx->thread_count; i++) {
2830 H264Context *c;
2831 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2832 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2833 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2834 c->h264dsp = h->h264dsp;
2835 c->sps = h->sps;
2836 c->pps = h->pps;
2837 c->pixel_shift = h->pixel_shift;
2838 init_scan_tables(c);
2839 clone_tables(c, h, i);
2840 }
2841
2842 for(i = 0; i < s->avctx->thread_count; i++)
2843 if (context_init(h->thread_context[i]) < 0) {
2844 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2845 return -1;
2846 }
2847 }
2848 }
2849
2850 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2851
2852 h->mb_mbaff = 0;
2853 h->mb_aff_frame = 0;
2854 last_pic_structure = s0->picture_structure;
2855 if(h->sps.frame_mbs_only_flag){
2856 s->picture_structure= PICT_FRAME;
2857 }else{
2858 if(get_bits1(&s->gb)) { //field_pic_flag
2859 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2860 } else {
2861 s->picture_structure= PICT_FRAME;
2862 h->mb_aff_frame = h->sps.mb_aff;
2863 }
2864 }
2865 h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2866
2867 if(h0->current_slice == 0){
2868 // Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
2869 if(h->frame_num != h->prev_frame_num) {
2870 int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
2871
2872 if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2873
2874 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2875 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2876 if (unwrap_prev_frame_num < 0)
2877 unwrap_prev_frame_num += max_frame_num;
2878
2879 h->prev_frame_num = unwrap_prev_frame_num;
2880 }
2881 }
2882
2883 while(h->frame_num != h->prev_frame_num &&
2884 h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
2885 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2886 av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
2887 if (ff_h264_frame_start(h) < 0)
2888 return -1;
2889 h->prev_frame_num++;
2890 h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2891 s->current_picture_ptr->frame_num= h->prev_frame_num;
2892 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
2893 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
2894 ff_generate_sliding_window_mmcos(h);
2895 if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2896 (s->avctx->err_recognition & AV_EF_EXPLODE))
2897 return AVERROR_INVALIDDATA;
2898 /* Error concealment: if a ref is missing, copy the previous ref in its place.
2899 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2900 * about there being no actual duplicates.
2901 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
2902 * concealing a lost frame, this probably isn't noticeable by comparison, but it should
2903 * be fixed. */
2904 if (h->short_ref_count) {
2905 if (prev) {
2906 av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2907 (const uint8_t**)prev->f.data, prev->f.linesize,
2908 s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2909 h->short_ref[0]->poc = prev->poc+2;
2910 }
2911 h->short_ref[0]->frame_num = h->prev_frame_num;
2912 }
2913 }
2914
2915 /* See if we have a decoded first field looking for a pair... */
2916 if (s0->first_field) {
2917 assert(s0->current_picture_ptr);
2918 assert(s0->current_picture_ptr->f.data[0]);
2919 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2920
2921 /* figure out if we have a complementary field pair */
2922 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2923 /*
2924 * Previous field is unmatched. Don't display it, but let it
2925 * remain for reference if marked as such.
2926 */
2927 s0->current_picture_ptr = NULL;
2928 s0->first_field = FIELD_PICTURE;
2929
2930 } else {
2931 if (h->nal_ref_idc &&
2932 s0->current_picture_ptr->f.reference &&
2933 s0->current_picture_ptr->frame_num != h->frame_num) {
2934 /*
2935 * This and previous field were reference, but had
2936 * different frame_nums. Consider this field first in
2937 * pair. Throw away previous field except for reference
2938 * purposes.
2939 */
2940 s0->first_field = 1;
2941 s0->current_picture_ptr = NULL;
2942
2943 } else {
2944 /* Second field in complementary pair */
2945 s0->first_field = 0;
2946 }
2947 }
2948
2949 } else {
2950 /* Frame or first field in a potentially complementary pair */
2951 assert(!s0->current_picture_ptr);
2952 s0->first_field = FIELD_PICTURE;
2953 }
2954
2955 if(!FIELD_PICTURE || s0->first_field) {
2956 if (ff_h264_frame_start(h) < 0) {
2957 s0->first_field = 0;
2958 return -1;
2959 }
2960 } else {
2961 ff_release_unused_pictures(s, 0);
2962 }
2963 }
2964 if(h != h0)
2965 clone_slice(h, h0);
2966
2967 s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2968
2969 assert(s->mb_num == s->mb_width * s->mb_height);
2970 if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2971 first_mb_in_slice >= s->mb_num){
2972 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2973 return -1;
2974 }
2975 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2976 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2977 if (s->picture_structure == PICT_BOTTOM_FIELD)
2978 s->resync_mb_y = s->mb_y = s->mb_y + 1;
2979 assert(s->mb_y < s->mb_height);
2980
2981 if(s->picture_structure==PICT_FRAME){
2982 h->curr_pic_num= h->frame_num;
2983 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2984 }else{
2985 h->curr_pic_num= 2*h->frame_num + 1;
2986 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2987 }
2988
2989 if(h->nal_unit_type == NAL_IDR_SLICE){
2990 get_ue_golomb(&s->gb); /* idr_pic_id */
2991 }
2992
2993 if(h->sps.poc_type==0){
2994 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2995
2996 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2997 h->delta_poc_bottom= get_se_golomb(&s->gb);
2998 }
2999 }
3000
3001 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3002 h->delta_poc[0]= get_se_golomb(&s->gb);
3003
3004 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3005 h->delta_poc[1]= get_se_golomb(&s->gb);
3006 }
3007
3008 init_poc(h);
3009
3010 if(h->pps.redundant_pic_cnt_present){
3011 h->redundant_pic_count= get_ue_golomb(&s->gb);
3012 }
3013
3014 //set defaults, might be overridden a few lines later
3015 h->ref_count[0]= h->pps.ref_count[0];
3016 h->ref_count[1]= h->pps.ref_count[1];
3017
3018 if(h->slice_type_nos != AV_PICTURE_TYPE_I){
3019 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
3020 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3021 }
3022 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3023
3024 if(num_ref_idx_active_override_flag){
3025 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3026 if(h->slice_type_nos==AV_PICTURE_TYPE_B)
3027 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3028
3029 if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
3030 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3031 h->ref_count[0]= h->ref_count[1]= 1;
3032 return -1;
3033 }
3034 }
3035 if(h->slice_type_nos == AV_PICTURE_TYPE_B)
3036 h->list_count= 2;
3037 else
3038 h->list_count= 1;
3039 }else
3040 h->list_count= 0;
3041
3042 if(!default_ref_list_done){
3043 ff_h264_fill_default_ref_list(h);
3044 }
3045
3046 if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0) {
3047 h->ref_count[1]= h->ref_count[0]= 0;
3048 return -1;
3049 }
3050
3051 if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
3052 s->last_picture_ptr= &h->ref_list[0][0];
3053 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3054 }
3055 if(h->slice_type_nos==AV_PICTURE_TYPE_B){
3056 s->next_picture_ptr= &h->ref_list[1][0];
3057 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3058 }
3059
3060 if( (h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P )
3061 || (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
3062 pred_weight_table(h);
3063 else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
3064 implicit_weight_table(h, -1);
3065 }else {
3066 h->use_weight = 0;
3067 for (i = 0; i < 2; i++) {
3068 h->luma_weight_flag[i] = 0;
3069 h->chroma_weight_flag[i] = 0;
3070 }
3071 }
3072
3073 if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
3074 (s->avctx->err_recognition & AV_EF_EXPLODE))
3075 return AVERROR_INVALIDDATA;
3076
3077 if(FRAME_MBAFF){
3078 ff_h264_fill_mbaff_ref_list(h);
3079
3080 if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
3081 implicit_weight_table(h, 0);
3082 implicit_weight_table(h, 1);
3083 }
3084 }
3085
3086 if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3087 ff_h264_direct_dist_scale_factor(h);
3088 ff_h264_direct_ref_list_init(h);
3089
3090 if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
3091 tmp = get_ue_golomb_31(&s->gb);
3092 if(tmp > 2){
3093 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3094 return -1;
3095 }
3096 h->cabac_init_idc= tmp;
3097 }
3098
3099 h->last_qscale_diff = 0;
3100 tmp = h->pps.init_qp +