CrystalHD: Remove redundant interlaced check.
[ffmpeg.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 FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21
22 /**
23  * @file
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[52]={
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,
50 };
51
52 static const uint8_t div6[52]={
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,
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_YUVJ420P,
60     PIX_FMT_NONE
61 };
62
63 void ff_h264_write_back_intra_pred_mode(H264Context *h){
64     int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
65
66     AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
67     mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
68     mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
69     mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
70 }
71
72 /**
73  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
74  */
75 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
76     MpegEncContext * const s = &h->s;
77     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
78     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
79     int i;
80
81     if(!(h->top_samples_available&0x8000)){
82         for(i=0; i<4; i++){
83             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
84             if(status<0){
85                 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);
86                 return -1;
87             } else if(status){
88                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
89             }
90         }
91     }
92
93     if((h->left_samples_available&0x8888)!=0x8888){
94         static const int mask[4]={0x8000,0x2000,0x80,0x20};
95         for(i=0; i<4; i++){
96             if(!(h->left_samples_available&mask[i])){
97                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
98                 if(status<0){
99                     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);
100                     return -1;
101                 } else if(status){
102                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
103                 }
104             }
105         }
106     }
107
108     return 0;
109 } //FIXME cleanup like ff_h264_check_intra_pred_mode
110
111 /**
112  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
113  */
114 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
115     MpegEncContext * const s = &h->s;
116     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
117     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
118
119     if(mode > 6U) {
120         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);
121         return -1;
122     }
123
124     if(!(h->top_samples_available&0x8000)){
125         mode= top[ mode ];
126         if(mode<0){
127             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);
128             return -1;
129         }
130     }
131
132     if((h->left_samples_available&0x8080) != 0x8080){
133         mode= left[ mode ];
134         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
135             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
136         }
137         if(mode<0){
138             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);
139             return -1;
140         }
141     }
142
143     return mode;
144 }
145
146 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
147     int i, si, di;
148     uint8_t *dst;
149     int bufidx;
150
151 //    src[0]&0x80;                //forbidden bit
152     h->nal_ref_idc= src[0]>>5;
153     h->nal_unit_type= src[0]&0x1F;
154
155     src++; length--;
156 #if 0
157     for(i=0; i<length; i++)
158         printf("%2X ", src[i]);
159 #endif
160
161 #if HAVE_FAST_UNALIGNED
162 # if HAVE_FAST_64BIT
163 #   define RS 7
164     for(i=0; i+1<length; i+=9){
165         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
166 # else
167 #   define RS 3
168     for(i=0; i+1<length; i+=5){
169         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
170 # endif
171             continue;
172         if(i>0 && !src[i]) i--;
173         while(src[i]) i++;
174 #else
175 #   define RS 0
176     for(i=0; i+1<length; i+=2){
177         if(src[i]) continue;
178         if(i>0 && src[i-1]==0) i--;
179 #endif
180         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
181             if(src[i+2]!=3){
182                 /* startcode, so we must be past the end */
183                 length=i;
184             }
185             break;
186         }
187         i-= RS;
188     }
189
190     if(i>=length-1){ //no escaped 0
191         *dst_length= length;
192         *consumed= length+1; //+1 for the header
193         return src;
194     }
195
196     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
197     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
198     dst= h->rbsp_buffer[bufidx];
199
200     if (dst == NULL){
201         return NULL;
202     }
203
204 //printf("decoding esc\n");
205     memcpy(dst, src, i);
206     si=di=i;
207     while(si+2<length){
208         //remove escapes (very rare 1:2^22)
209         if(src[si+2]>3){
210             dst[di++]= src[si++];
211             dst[di++]= src[si++];
212         }else if(src[si]==0 && src[si+1]==0){
213             if(src[si+2]==3){ //escape
214                 dst[di++]= 0;
215                 dst[di++]= 0;
216                 si+=3;
217                 continue;
218             }else //next start code
219                 goto nsc;
220         }
221
222         dst[di++]= src[si++];
223     }
224     while(si<length)
225         dst[di++]= src[si++];
226 nsc:
227
228     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
229
230     *dst_length= di;
231     *consumed= si + 1;//+1 for the header
232 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
233     return dst;
234 }
235
236 /**
237  * Identify the exact end of the bitstream
238  * @return the length of the trailing, or 0 if damaged
239  */
240 static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
241     int v= *src;
242     int r;
243
244     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
245
246     for(r=1; r<9; r++){
247         if(v&1) return r;
248         v>>=1;
249     }
250     return 0;
251 }
252
253 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height,
254                                  int y_offset, int list){
255     int raw_my= h->mv_cache[list][ scan8[n] ][1];
256     int filter_height= (raw_my&3) ? 2 : 0;
257     int full_my= (raw_my>>2) + y_offset;
258     int top = full_my - filter_height, bottom = full_my + height + filter_height;
259
260     return FFMAX(abs(top), bottom);
261 }
262
263 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height,
264                                int y_offset, int list0, int list1, int *nrefs){
265     MpegEncContext * const s = &h->s;
266     int my;
267
268     y_offset += 16*(s->mb_y >> MB_FIELD);
269
270     if(list0){
271         int ref_n = h->ref_cache[0][ scan8[n] ];
272         Picture *ref= &h->ref_list[0][ref_n];
273
274         // Error resilience puts the current picture in the ref list.
275         // Don't try to wait on these as it will cause a deadlock.
276         // Fields can wait on each other, though.
277         if(ref->thread_opaque != s->current_picture.thread_opaque ||
278            (ref->reference&3) != s->picture_structure) {
279             my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
280             if (refs[0][ref_n] < 0) nrefs[0] += 1;
281             refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
282         }
283     }
284
285     if(list1){
286         int ref_n = h->ref_cache[1][ scan8[n] ];
287         Picture *ref= &h->ref_list[1][ref_n];
288
289         if(ref->thread_opaque != s->current_picture.thread_opaque ||
290            (ref->reference&3) != s->picture_structure) {
291             my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
292             if (refs[1][ref_n] < 0) nrefs[1] += 1;
293             refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
294         }
295     }
296 }
297
298 /**
299  * Wait until all reference frames are available for MC operations.
300  *
301  * @param h the H264 context
302  */
303 static void await_references(H264Context *h){
304     MpegEncContext * const s = &h->s;
305     const int mb_xy= h->mb_xy;
306     const int mb_type= s->current_picture.mb_type[mb_xy];
307     int refs[2][48];
308     int nrefs[2] = {0};
309     int ref, list;
310
311     memset(refs, -1, sizeof(refs));
312
313     if(IS_16X16(mb_type)){
314         get_lowest_part_y(h, refs, 0, 16, 0,
315                   IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
316     }else if(IS_16X8(mb_type)){
317         get_lowest_part_y(h, refs, 0, 8, 0,
318                   IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
319         get_lowest_part_y(h, refs, 8, 8, 8,
320                   IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
321     }else if(IS_8X16(mb_type)){
322         get_lowest_part_y(h, refs, 0, 16, 0,
323                   IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
324         get_lowest_part_y(h, refs, 4, 16, 0,
325                   IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
326     }else{
327         int i;
328
329         assert(IS_8X8(mb_type));
330
331         for(i=0; i<4; i++){
332             const int sub_mb_type= h->sub_mb_type[i];
333             const int n= 4*i;
334             int y_offset= (i&2)<<2;
335
336             if(IS_SUB_8X8(sub_mb_type)){
337                 get_lowest_part_y(h, refs, n  , 8, y_offset,
338                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
339             }else if(IS_SUB_8X4(sub_mb_type)){
340                 get_lowest_part_y(h, refs, n  , 4, y_offset,
341                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
342                 get_lowest_part_y(h, refs, n+2, 4, y_offset+4,
343                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
344             }else if(IS_SUB_4X8(sub_mb_type)){
345                 get_lowest_part_y(h, refs, n  , 8, y_offset,
346                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
347                 get_lowest_part_y(h, refs, n+1, 8, y_offset,
348                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
349             }else{
350                 int j;
351                 assert(IS_SUB_4X4(sub_mb_type));
352                 for(j=0; j<4; j++){
353                     int sub_y_offset= y_offset + 2*(j&2);
354                     get_lowest_part_y(h, refs, n+j, 4, sub_y_offset,
355                               IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
356                 }
357             }
358         }
359     }
360
361     for(list=h->list_count-1; list>=0; list--){
362         for(ref=0; ref<48 && nrefs[list]; ref++){
363             int row = refs[list][ref];
364             if(row >= 0){
365                 Picture *ref_pic = &h->ref_list[list][ref];
366                 int ref_field = ref_pic->reference - 1;
367                 int ref_field_picture = ref_pic->field_picture;
368                 int pic_height = 16*s->mb_height >> ref_field_picture;
369
370                 row <<= MB_MBAFF;
371                 nrefs[list]--;
372
373                 if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields
374                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) - !(row&1), pic_height-1), 1);
375                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1)           , pic_height-1), 0);
376                 }else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame
377                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row*2 + ref_field    , pic_height-1), 0);
378                 }else if(FIELD_PICTURE){
379                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), ref_field);
380                 }else{
381                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), 0);
382                 }
383             }
384         }
385     }
386 }
387
388 #if 0
389 /**
390  * DCT transforms the 16 dc values.
391  * @param qp quantization parameter ??? FIXME
392  */
393 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
394 //    const int qmul= dequant_coeff[qp][0];
395     int i;
396     int temp[16]; //FIXME check if this is a good idea
397     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
398     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
399
400     for(i=0; i<4; i++){
401         const int offset= y_offset[i];
402         const int z0= block[offset+stride*0] + block[offset+stride*4];
403         const int z1= block[offset+stride*0] - block[offset+stride*4];
404         const int z2= block[offset+stride*1] - block[offset+stride*5];
405         const int z3= block[offset+stride*1] + block[offset+stride*5];
406
407         temp[4*i+0]= z0+z3;
408         temp[4*i+1]= z1+z2;
409         temp[4*i+2]= z1-z2;
410         temp[4*i+3]= z0-z3;
411     }
412
413     for(i=0; i<4; i++){
414         const int offset= x_offset[i];
415         const int z0= temp[4*0+i] + temp[4*2+i];
416         const int z1= temp[4*0+i] - temp[4*2+i];
417         const int z2= temp[4*1+i] - temp[4*3+i];
418         const int z3= temp[4*1+i] + temp[4*3+i];
419
420         block[stride*0 +offset]= (z0 + z3)>>1;
421         block[stride*2 +offset]= (z1 + z2)>>1;
422         block[stride*8 +offset]= (z1 - z2)>>1;
423         block[stride*10+offset]= (z0 - z3)>>1;
424     }
425 }
426 #endif
427
428 #undef xStride
429 #undef stride
430
431 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qmul){
432     const int stride= 16*2;
433     const int xStride= 16;
434     int a,b,c,d,e;
435
436     a= block[stride*0 + xStride*0];
437     b= block[stride*0 + xStride*1];
438     c= block[stride*1 + xStride*0];
439     d= block[stride*1 + xStride*1];
440
441     e= a-b;
442     a= a+b;
443     b= c-d;
444     c= c+d;
445
446     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
447     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
448     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
449     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
450 }
451
452 #if 0
453 static void chroma_dc_dct_c(DCTELEM *block){
454     const int stride= 16*2;
455     const int xStride= 16;
456     int a,b,c,d,e;
457
458     a= block[stride*0 + xStride*0];
459     b= block[stride*0 + xStride*1];
460     c= block[stride*1 + xStride*0];
461     d= block[stride*1 + xStride*1];
462
463     e= a-b;
464     a= a+b;
465     b= c-d;
466     c= c+d;
467
468     block[stride*0 + xStride*0]= (a+c);
469     block[stride*0 + xStride*1]= (e+b);
470     block[stride*1 + xStride*0]= (a-c);
471     block[stride*1 + xStride*1]= (e-b);
472 }
473 #endif
474
475 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
476                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
477                            int src_x_offset, int src_y_offset,
478                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
479     MpegEncContext * const s = &h->s;
480     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
481     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
482     const int luma_xy= (mx&3) + ((my&3)<<2);
483     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
484     uint8_t * src_cb, * src_cr;
485     int extra_width= h->emu_edge_width;
486     int extra_height= h->emu_edge_height;
487     int emu=0;
488     const int full_mx= mx>>2;
489     const int full_my= my>>2;
490     const int pic_width  = 16*s->mb_width;
491     const int pic_height = 16*s->mb_height >> MB_FIELD;
492
493     if(mx&7) extra_width -= 3;
494     if(my&7) extra_height -= 3;
495
496     if(   full_mx < 0-extra_width
497        || full_my < 0-extra_height
498        || full_mx + 16/*FIXME*/ > pic_width + extra_width
499        || full_my + 16/*FIXME*/ > pic_height + extra_height){
500         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
501             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
502         emu=1;
503     }
504
505     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
506     if(!square){
507         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
508     }
509
510     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
511
512     if(MB_FIELD){
513         // chroma offset when predicting from a field of opposite parity
514         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
515         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
516     }
517     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
518     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
519
520     if(emu){
521         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
522             src_cb= s->edge_emu_buffer;
523     }
524     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
525
526     if(emu){
527         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
528             src_cr= s->edge_emu_buffer;
529     }
530     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
531 }
532
533 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
534                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
535                            int x_offset, int y_offset,
536                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
537                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
538                            int list0, int list1){
539     MpegEncContext * const s = &h->s;
540     qpel_mc_func *qpix_op=  qpix_put;
541     h264_chroma_mc_func chroma_op= chroma_put;
542
543     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
544     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
545     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
546     x_offset += 8*s->mb_x;
547     y_offset += 8*(s->mb_y >> MB_FIELD);
548
549     if(list0){
550         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
551         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
552                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
553                            qpix_op, chroma_op);
554
555         qpix_op=  qpix_avg;
556         chroma_op= chroma_avg;
557     }
558
559     if(list1){
560         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
561         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
562                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
563                            qpix_op, chroma_op);
564     }
565 }
566
567 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
568                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
569                            int x_offset, int y_offset,
570                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
571                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
572                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
573                            int list0, int list1){
574     MpegEncContext * const s = &h->s;
575
576     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
577     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
578     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
579     x_offset += 8*s->mb_x;
580     y_offset += 8*(s->mb_y >> MB_FIELD);
581
582     if(list0 && list1){
583         /* don't optimize for luma-only case, since B-frames usually
584          * use implicit weights => chroma too. */
585         uint8_t *tmp_cb = s->obmc_scratchpad;
586         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
587         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
588         int refn0 = h->ref_cache[0][ scan8[n] ];
589         int refn1 = h->ref_cache[1][ scan8[n] ];
590
591         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
592                     dest_y, dest_cb, dest_cr,
593                     x_offset, y_offset, qpix_put, chroma_put);
594         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
595                     tmp_y, tmp_cb, tmp_cr,
596                     x_offset, y_offset, qpix_put, chroma_put);
597
598         if(h->use_weight == 2){
599             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
600             int weight1 = 64 - weight0;
601             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
602             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
603             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
604         }else{
605             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
606                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
607                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
608             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
609                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
610                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
611             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
612                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
613                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
614         }
615     }else{
616         int list = list1 ? 1 : 0;
617         int refn = h->ref_cache[list][ scan8[n] ];
618         Picture *ref= &h->ref_list[list][refn];
619         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
620                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
621                     qpix_put, chroma_put);
622
623         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
624                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
625         if(h->use_weight_chroma){
626             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
627                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
628             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
629                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
630         }
631     }
632 }
633
634 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
635                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
636                            int x_offset, int y_offset,
637                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
638                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
639                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
640                            int list0, int list1){
641     if((h->use_weight==2 && list0 && list1
642         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
643        || h->use_weight==1)
644         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
645                          x_offset, y_offset, qpix_put, chroma_put,
646                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
647     else
648         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
649                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
650 }
651
652 static inline void prefetch_motion(H264Context *h, int list){
653     /* fetch pixels for estimated mv 4 macroblocks ahead
654      * optimized for 64byte cache lines */
655     MpegEncContext * const s = &h->s;
656     const int refn = h->ref_cache[list][scan8[0]];
657     if(refn >= 0){
658         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
659         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
660         uint8_t **src= h->ref_list[list][refn].data;
661         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
662         s->dsp.prefetch(src[0]+off, s->linesize, 4);
663         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
664         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
665     }
666 }
667
668 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
669                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
670                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
671                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
672     MpegEncContext * const s = &h->s;
673     const int mb_xy= h->mb_xy;
674     const int mb_type= s->current_picture.mb_type[mb_xy];
675
676     assert(IS_INTER(mb_type));
677
678     if(HAVE_PTHREADS && s->avctx->active_thread_type&FF_THREAD_FRAME)
679         await_references(h);
680     prefetch_motion(h, 0);
681
682     if(IS_16X16(mb_type)){
683         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
684                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
685                 weight_op, weight_avg,
686                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
687     }else if(IS_16X8(mb_type)){
688         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
689                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
690                 &weight_op[1], &weight_avg[1],
691                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
692         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
693                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
694                 &weight_op[1], &weight_avg[1],
695                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
696     }else if(IS_8X16(mb_type)){
697         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
698                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
699                 &weight_op[2], &weight_avg[2],
700                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
701         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
702                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
703                 &weight_op[2], &weight_avg[2],
704                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
705     }else{
706         int i;
707
708         assert(IS_8X8(mb_type));
709
710         for(i=0; i<4; i++){
711             const int sub_mb_type= h->sub_mb_type[i];
712             const int n= 4*i;
713             int x_offset= (i&1)<<2;
714             int y_offset= (i&2)<<1;
715
716             if(IS_SUB_8X8(sub_mb_type)){
717                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
718                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
719                     &weight_op[3], &weight_avg[3],
720                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
721             }else if(IS_SUB_8X4(sub_mb_type)){
722                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
723                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
724                     &weight_op[4], &weight_avg[4],
725                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
726                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
727                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
728                     &weight_op[4], &weight_avg[4],
729                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
730             }else if(IS_SUB_4X8(sub_mb_type)){
731                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
732                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
733                     &weight_op[5], &weight_avg[5],
734                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
735                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
736                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
737                     &weight_op[5], &weight_avg[5],
738                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
739             }else{
740                 int j;
741                 assert(IS_SUB_4X4(sub_mb_type));
742                 for(j=0; j<4; j++){
743                     int sub_x_offset= x_offset + 2*(j&1);
744                     int sub_y_offset= y_offset +   (j&2);
745                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
746                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
747                         &weight_op[6], &weight_avg[6],
748                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
749                 }
750             }
751         }
752     }
753
754     prefetch_motion(h, 1);
755 }
756
757
758 static void free_tables(H264Context *h, int free_rbsp){
759     int i;
760     H264Context *hx;
761     av_freep(&h->intra4x4_pred_mode);
762     av_freep(&h->chroma_pred_mode_table);
763     av_freep(&h->cbp_table);
764     av_freep(&h->mvd_table[0]);
765     av_freep(&h->mvd_table[1]);
766     av_freep(&h->direct_table);
767     av_freep(&h->non_zero_count);
768     av_freep(&h->slice_table_base);
769     h->slice_table= NULL;
770     av_freep(&h->list_counts);
771
772     av_freep(&h->mb2b_xy);
773     av_freep(&h->mb2br_xy);
774
775     for(i = 0; i < MAX_THREADS; i++) {
776         hx = h->thread_context[i];
777         if(!hx) continue;
778         av_freep(&hx->top_borders[1]);
779         av_freep(&hx->top_borders[0]);
780         av_freep(&hx->s.obmc_scratchpad);
781         if (free_rbsp){
782             av_freep(&hx->rbsp_buffer[1]);
783             av_freep(&hx->rbsp_buffer[0]);
784             hx->rbsp_buffer_size[0] = 0;
785             hx->rbsp_buffer_size[1] = 0;
786         }
787         if (i) av_freep(&h->thread_context[i]);
788     }
789 }
790
791 static void init_dequant8_coeff_table(H264Context *h){
792     int i,q,x;
793     h->dequant8_coeff[0] = h->dequant8_buffer[0];
794     h->dequant8_coeff[1] = h->dequant8_buffer[1];
795
796     for(i=0; i<2; i++ ){
797         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
798             h->dequant8_coeff[1] = h->dequant8_buffer[0];
799             break;
800         }
801
802         for(q=0; q<52; q++){
803             int shift = div6[q];
804             int idx = rem6[q];
805             for(x=0; x<64; x++)
806                 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
807                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
808                     h->pps.scaling_matrix8[i][x]) << shift;
809         }
810     }
811 }
812
813 static void init_dequant4_coeff_table(H264Context *h){
814     int i,j,q,x;
815     for(i=0; i<6; i++ ){
816         h->dequant4_coeff[i] = h->dequant4_buffer[i];
817         for(j=0; j<i; j++){
818             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
819                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
820                 break;
821             }
822         }
823         if(j<i)
824             continue;
825
826         for(q=0; q<52; q++){
827             int shift = div6[q] + 2;
828             int idx = rem6[q];
829             for(x=0; x<16; x++)
830                 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
831                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
832                     h->pps.scaling_matrix4[i][x]) << shift;
833         }
834     }
835 }
836
837 static void init_dequant_tables(H264Context *h){
838     int i,x;
839     init_dequant4_coeff_table(h);
840     if(h->pps.transform_8x8_mode)
841         init_dequant8_coeff_table(h);
842     if(h->sps.transform_bypass){
843         for(i=0; i<6; i++)
844             for(x=0; x<16; x++)
845                 h->dequant4_coeff[i][0][x] = 1<<6;
846         if(h->pps.transform_8x8_mode)
847             for(i=0; i<2; i++)
848                 for(x=0; x<64; x++)
849                     h->dequant8_coeff[i][0][x] = 1<<6;
850     }
851 }
852
853
854 int ff_h264_alloc_tables(H264Context *h){
855     MpegEncContext * const s = &h->s;
856     const int big_mb_num= s->mb_stride * (s->mb_height+1);
857     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
858     int x,y;
859
860     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
861
862     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
863     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
864     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
865
866     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
867     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
868     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
869     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
870     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
871
872     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
873     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
874
875     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
876     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
877     for(y=0; y<s->mb_height; y++){
878         for(x=0; x<s->mb_width; x++){
879             const int mb_xy= x + y*s->mb_stride;
880             const int b_xy = 4*x + 4*y*h->b_stride;
881
882             h->mb2b_xy [mb_xy]= b_xy;
883             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
884         }
885     }
886
887     s->obmc_scratchpad = NULL;
888
889     if(!h->dequant4_coeff[0])
890         init_dequant_tables(h);
891
892     return 0;
893 fail:
894     free_tables(h, 1);
895     return -1;
896 }
897
898 /**
899  * Mimic alloc_tables(), but for every context thread.
900  */
901 static void clone_tables(H264Context *dst, H264Context *src, int i){
902     MpegEncContext * const s = &src->s;
903     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
904     dst->non_zero_count           = src->non_zero_count;
905     dst->slice_table              = src->slice_table;
906     dst->cbp_table                = src->cbp_table;
907     dst->mb2b_xy                  = src->mb2b_xy;
908     dst->mb2br_xy                 = src->mb2br_xy;
909     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
910     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
911     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
912     dst->direct_table             = src->direct_table;
913     dst->list_counts              = src->list_counts;
914
915     dst->s.obmc_scratchpad = NULL;
916     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
917 }
918
919 /**
920  * Init context
921  * Allocate buffers which are not shared amongst multiple threads.
922  */
923 static int context_init(H264Context *h){
924     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
925     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
926
927     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
928     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
929
930     return 0;
931 fail:
932     return -1; // free_tables will clean up for us
933 }
934
935 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
936
937 static av_cold void common_init(H264Context *h){
938     MpegEncContext * const s = &h->s;
939
940     s->width = s->avctx->width;
941     s->height = s->avctx->height;
942     s->codec_id= s->avctx->codec->id;
943
944     ff_h264dsp_init(&h->h264dsp);
945     ff_h264_pred_init(&h->hpc, s->codec_id);
946
947     h->dequant_coeff_pps= -1;
948     s->unrestricted_mv=1;
949     s->decode=1; //FIXME
950
951     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
952
953     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
954     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
955 }
956
957 int ff_h264_decode_extradata(H264Context *h)
958 {
959     AVCodecContext *avctx = h->s.avctx;
960
961     if(*(char *)avctx->extradata == 1){
962         int i, cnt, nalsize;
963         unsigned char *p = avctx->extradata;
964
965         h->is_avc = 1;
966
967         if(avctx->extradata_size < 7) {
968             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
969             return -1;
970         }
971         /* sps and pps in the avcC always have length coded with 2 bytes,
972            so put a fake nal_length_size = 2 while parsing them */
973         h->nal_length_size = 2;
974         // Decode sps from avcC
975         cnt = *(p+5) & 0x1f; // Number of sps
976         p += 6;
977         for (i = 0; i < cnt; i++) {
978             nalsize = AV_RB16(p) + 2;
979             if(decode_nal_units(h, p, nalsize) < 0) {
980                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
981                 return -1;
982             }
983             p += nalsize;
984         }
985         // Decode pps from avcC
986         cnt = *(p++); // Number of pps
987         for (i = 0; i < cnt; i++) {
988             nalsize = AV_RB16(p) + 2;
989             if(decode_nal_units(h, p, nalsize)  != nalsize) {
990                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
991                 return -1;
992             }
993             p += nalsize;
994         }
995         // Now store right nal length size, that will be use to parse all other nals
996         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
997     } else {
998         h->is_avc = 0;
999         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
1000             return -1;
1001     }
1002     return 0;
1003 }
1004
1005 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1006     H264Context *h= avctx->priv_data;
1007     MpegEncContext * const s = &h->s;
1008
1009     MPV_decode_defaults(s);
1010
1011     s->avctx = avctx;
1012     common_init(h);
1013
1014     s->out_format = FMT_H264;
1015     s->workaround_bugs= avctx->workaround_bugs;
1016
1017     // set defaults
1018 //    s->decode_mb= ff_h263_decode_mb;
1019     s->quarter_sample = 1;
1020     if(!avctx->has_b_frames)
1021     s->low_delay= 1;
1022
1023     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1024
1025     ff_h264_decode_init_vlc();
1026
1027     h->thread_context[0] = h;
1028     h->outputed_poc = h->next_outputed_poc = INT_MIN;
1029     h->prev_poc_msb= 1<<16;
1030     h->x264_build = -1;
1031     ff_h264_reset_sei(h);
1032     if(avctx->codec_id == CODEC_ID_H264){
1033         if(avctx->ticks_per_frame == 1){
1034             s->avctx->time_base.den *=2;
1035         }
1036         avctx->ticks_per_frame = 2;
1037     }
1038
1039     if(avctx->extradata_size > 0 && avctx->extradata &&
1040         ff_h264_decode_extradata(h))
1041         return -1;
1042
1043     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1044         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1045         s->low_delay = 0;
1046     }
1047
1048     return 0;
1049 }
1050
1051 static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1052 {
1053     int i;
1054
1055     for (i=0; i<count; i++){
1056         to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1057     }
1058 }
1059
1060 static void copy_parameter_set(void **to, void **from, int count, int size)
1061 {
1062     int i;
1063
1064     for (i=0; i<count; i++){
1065         if (to[i] && !from[i]) av_freep(&to[i]);
1066         else if (from[i] && !to[i]) to[i] = av_malloc(size);
1067
1068         if (from[i]) memcpy(to[i], from[i], size);
1069     }
1070 }
1071
1072 static int decode_init_thread_copy(AVCodecContext *avctx){
1073     H264Context *h= avctx->priv_data;
1074
1075     if (!avctx->is_copy) return 0;
1076     memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1077     memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1078
1079     return 0;
1080 }
1081
1082 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1083 static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1084     H264Context *h= dst->priv_data, *h1= src->priv_data;
1085     MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1086     int inited = s->context_initialized, err;
1087     int i;
1088
1089     if(dst == src || !s1->context_initialized) return 0;
1090
1091     err = ff_mpeg_update_thread_context(dst, src);
1092     if(err) return err;
1093
1094     //FIXME handle width/height changing
1095     if(!inited){
1096         for(i = 0; i < MAX_SPS_COUNT; i++)
1097             av_freep(h->sps_buffers + i);
1098
1099         for(i = 0; i < MAX_PPS_COUNT; i++)
1100             av_freep(h->pps_buffers + i);
1101
1102         memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1103         memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1104         memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1105         ff_h264_alloc_tables(h);
1106         context_init(h);
1107
1108         for(i=0; i<2; i++){
1109             h->rbsp_buffer[i] = NULL;
1110             h->rbsp_buffer_size[i] = 0;
1111         }
1112
1113         h->thread_context[0] = h;
1114
1115         // frame_start may not be called for the next thread (if it's decoding a bottom field)
1116         // so this has to be allocated here
1117         h->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
1118
1119         s->dsp.clear_blocks(h->mb);
1120     }
1121
1122     //extradata/NAL handling
1123     h->is_avc          = h1->is_avc;
1124
1125     //SPS/PPS
1126     copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1127     h->sps             = h1->sps;
1128     copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1129     h->pps             = h1->pps;
1130
1131     //Dequantization matrices
1132     //FIXME these are big - can they be only copied when PPS changes?
1133     copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1134
1135     for(i=0; i<6; i++)
1136         h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1137
1138     for(i=0; i<2; i++)
1139         h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1140
1141     h->dequant_coeff_pps = h1->dequant_coeff_pps;
1142
1143     //POC timing
1144     copy_fields(h, h1, poc_lsb, redundant_pic_count);
1145
1146     //reference lists
1147     copy_fields(h, h1, ref_count, intra_gb);
1148     copy_fields(h, h1, short_ref, cabac_init_idc);
1149
1150     copy_picture_range(h->short_ref,   h1->short_ref,   32, s, s1);
1151     copy_picture_range(h->long_ref,    h1->long_ref,    32,  s, s1);
1152     copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1153
1154     h->last_slice_type = h1->last_slice_type;
1155
1156     if(!s->current_picture_ptr) return 0;
1157
1158     if(!s->dropable) {
1159         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1160         h->prev_poc_msb     = h->poc_msb;
1161         h->prev_poc_lsb     = h->poc_lsb;
1162     }
1163     h->prev_frame_num_offset= h->frame_num_offset;
1164     h->prev_frame_num       = h->frame_num;
1165     h->outputed_poc         = h->next_outputed_poc;
1166
1167     return 0;
1168 }
1169
1170 int ff_h264_frame_start(H264Context *h){
1171     MpegEncContext * const s = &h->s;
1172     int i;
1173
1174     if(MPV_frame_start(s, s->avctx) < 0)
1175         return -1;
1176     ff_er_frame_start(s);
1177     /*
1178      * MPV_frame_start uses pict_type to derive key_frame.
1179      * This is incorrect for H.264; IDR markings must be used.
1180      * Zero here; IDR markings per slice in frame or fields are ORed in later.
1181      * See decode_nal_units().
1182      */
1183     s->current_picture_ptr->key_frame= 0;
1184     s->current_picture_ptr->mmco_reset= 0;
1185
1186     assert(s->linesize && s->uvlinesize);
1187
1188     for(i=0; i<16; i++){
1189         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
1190         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
1191     }
1192     for(i=0; i<4; i++){
1193         h->block_offset[16+i]=
1194         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1195         h->block_offset[24+16+i]=
1196         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1197     }
1198
1199     /* can't be in alloc_tables because linesize isn't known there.
1200      * FIXME: redo bipred weight to not require extra buffer? */
1201     for(i = 0; i < s->avctx->thread_count; i++)
1202         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1203             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
1204
1205     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
1206     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
1207
1208 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
1209
1210     // We mark the current picture as non-reference after allocating it, so
1211     // that if we break out due to an error it can be released automatically
1212     // in the next MPV_frame_start().
1213     // SVQ3 as well as most other codecs have only last/next/current and thus
1214     // get released even with set reference, besides SVQ3 and others do not
1215     // mark frames as reference later "naturally".
1216     if(s->codec_id != CODEC_ID_SVQ3)
1217         s->current_picture_ptr->reference= 0;
1218
1219     s->current_picture_ptr->field_poc[0]=
1220     s->current_picture_ptr->field_poc[1]= INT_MAX;
1221
1222     h->next_output_pic = NULL;
1223
1224     assert(s->current_picture_ptr->long_ref==0);
1225
1226     return 0;
1227 }
1228
1229 /**
1230   * Run setup operations that must be run after slice header decoding.
1231   * This includes finding the next displayed frame.
1232   *
1233   * @param h h264 master context
1234   */
1235 static void decode_postinit(H264Context *h){
1236     MpegEncContext * const s = &h->s;
1237     Picture *out = s->current_picture_ptr;
1238     Picture *cur = s->current_picture_ptr;
1239     int i, pics, out_of_order, out_idx;
1240
1241     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1242     s->current_picture_ptr->pict_type= s->pict_type;
1243
1244     if (h->next_output_pic) return;
1245
1246     if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1247         //FIXME this allows the next thread to start once we encounter the first field of a PAFF packet
1248         //This works if the next packet contains the second field. It does not work if both fields are
1249         //in the same packet.
1250         //ff_thread_finish_setup(s->avctx);
1251         return;
1252     }
1253
1254     cur->interlaced_frame = 0;
1255     cur->repeat_pict = 0;
1256
1257     /* Signal interlacing information externally. */
1258     /* Prioritize picture timing SEI information over used decoding process if it exists. */
1259
1260     if(h->sps.pic_struct_present_flag){
1261         switch (h->sei_pic_struct)
1262         {
1263         case SEI_PIC_STRUCT_FRAME:
1264             break;
1265         case SEI_PIC_STRUCT_TOP_FIELD:
1266         case SEI_PIC_STRUCT_BOTTOM_FIELD:
1267             cur->interlaced_frame = 1;
1268             break;
1269         case SEI_PIC_STRUCT_TOP_BOTTOM:
1270         case SEI_PIC_STRUCT_BOTTOM_TOP:
1271             if (FIELD_OR_MBAFF_PICTURE)
1272                 cur->interlaced_frame = 1;
1273             else
1274                 // try to flag soft telecine progressive
1275                 cur->interlaced_frame = h->prev_interlaced_frame;
1276             break;
1277         case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1278         case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1279             // Signal the possibility of telecined film externally (pic_struct 5,6)
1280             // From these hints, let the applications decide if they apply deinterlacing.
1281             cur->repeat_pict = 1;
1282             break;
1283         case SEI_PIC_STRUCT_FRAME_DOUBLING:
1284             // Force progressive here, as doubling interlaced frame is a bad idea.
1285             cur->repeat_pict = 2;
1286             break;
1287         case SEI_PIC_STRUCT_FRAME_TRIPLING:
1288             cur->repeat_pict = 4;
1289             break;
1290         }
1291
1292         if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1293             cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
1294     }else{
1295         /* Derive interlacing flag from used decoding process. */
1296         cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1297     }
1298     h->prev_interlaced_frame = cur->interlaced_frame;
1299
1300     if (cur->field_poc[0] != cur->field_poc[1]){
1301         /* Derive top_field_first from field pocs. */
1302         cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
1303     }else{
1304         if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
1305             /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1306             if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1307               || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1308                 cur->top_field_first = 1;
1309             else
1310                 cur->top_field_first = 0;
1311         }else{
1312             /* Most likely progressive */
1313             cur->top_field_first = 0;
1314         }
1315     }
1316
1317     //FIXME do something with unavailable reference frames
1318
1319     /* Sort B-frames into display order */
1320
1321     if(h->sps.bitstream_restriction_flag
1322        && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1323         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1324         s->low_delay = 0;
1325     }
1326
1327     if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1328        && !h->sps.bitstream_restriction_flag){
1329         s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
1330         s->low_delay= 0;
1331     }
1332
1333     pics = 0;
1334     while(h->delayed_pic[pics]) pics++;
1335
1336     assert(pics <= MAX_DELAYED_PIC_COUNT);
1337
1338     h->delayed_pic[pics++] = cur;
1339     if(cur->reference == 0)
1340         cur->reference = DELAYED_PIC_REF;
1341
1342     out = h->delayed_pic[0];
1343     out_idx = 0;
1344     for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
1345         if(h->delayed_pic[i]->poc < out->poc){
1346             out = h->delayed_pic[i];
1347             out_idx = i;
1348         }
1349     if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
1350         h->next_outputed_poc= INT_MIN;
1351     out_of_order = out->poc < h->next_outputed_poc;
1352
1353     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
1354         { }
1355     else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
1356        || (s->low_delay &&
1357         ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2)
1358          || cur->pict_type == FF_B_TYPE)))
1359     {
1360         s->low_delay = 0;
1361         s->avctx->has_b_frames++;
1362     }
1363
1364     if(out_of_order || pics > s->avctx->has_b_frames){
1365         out->reference &= ~DELAYED_PIC_REF;
1366         for(i=out_idx; h->delayed_pic[i]; i++)
1367             h->delayed_pic[i] = h->delayed_pic[i+1];
1368     }
1369     if(!out_of_order && pics > s->avctx->has_b_frames){
1370         h->next_output_pic = out;
1371         if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
1372             h->next_outputed_poc = INT_MIN;
1373         } else
1374             h->next_outputed_poc = out->poc;
1375     }else{
1376         av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1377     }
1378
1379     ff_thread_finish_setup(s->avctx);
1380 }
1381
1382 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
1383     MpegEncContext * const s = &h->s;
1384     uint8_t *top_border;
1385     int top_idx = 1;
1386
1387     src_y  -=   linesize;
1388     src_cb -= uvlinesize;
1389     src_cr -= uvlinesize;
1390
1391     if(!simple && FRAME_MBAFF){
1392         if(s->mb_y&1){
1393             if(!MB_MBAFF){
1394                 top_border = h->top_borders[0][s->mb_x];
1395                 AV_COPY128(top_border, src_y + 15*linesize);
1396                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1397                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1398                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1399                 }
1400             }
1401         }else if(MB_MBAFF){
1402             top_idx = 0;
1403         }else
1404             return;
1405     }
1406
1407     top_border = h->top_borders[top_idx][s->mb_x];
1408     // There are two lines saved, the line above the the top macroblock of a pair,
1409     // and the line above the bottom macroblock
1410     AV_COPY128(top_border, src_y + 16*linesize);
1411
1412     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1413         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1414         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1415     }
1416 }
1417
1418 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
1419     MpegEncContext * const s = &h->s;
1420     int deblock_left;
1421     int deblock_top;
1422     int top_idx = 1;
1423     uint8_t *top_border_m1;
1424     uint8_t *top_border;
1425
1426     if(!simple && FRAME_MBAFF){
1427         if(s->mb_y&1){
1428             if(!MB_MBAFF)
1429                 return;
1430         }else{
1431             top_idx = MB_MBAFF ? 0 : 1;
1432         }
1433     }
1434
1435     if(h->deblocking_filter == 2) {
1436         deblock_left = h->left_type[0];
1437         deblock_top  = h->top_type;
1438     } else {
1439         deblock_left = (s->mb_x > 0);
1440         deblock_top =  (s->mb_y > !!MB_FIELD);
1441     }
1442
1443     src_y  -=   linesize + 1;
1444     src_cb -= uvlinesize + 1;
1445     src_cr -= uvlinesize + 1;
1446
1447     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1448     top_border    = h->top_borders[top_idx][s->mb_x];
1449
1450 #define XCHG(a,b,xchg)\
1451 if (xchg) AV_SWAP64(b,a);\
1452 else      AV_COPY64(b,a);
1453
1454     if(deblock_top){
1455         if(deblock_left){
1456             XCHG(top_border_m1+8, src_y -7, 1);
1457         }
1458         XCHG(top_border+0, src_y +1, xchg);
1459         XCHG(top_border+8, src_y +9, 1);
1460         if(s->mb_x+1 < s->mb_width){
1461             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1462         }
1463     }
1464
1465     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1466         if(deblock_top){
1467             if(deblock_left){
1468                 XCHG(top_border_m1+16, src_cb -7, 1);
1469                 XCHG(top_border_m1+24, src_cr -7, 1);
1470             }
1471             XCHG(top_border+16, src_cb+1, 1);
1472             XCHG(top_border+24, src_cr+1, 1);
1473         }
1474     }
1475 }
1476
1477 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1478     MpegEncContext * const s = &h->s;
1479     const int mb_x= s->mb_x;
1480     const int mb_y= s->mb_y;
1481     const int mb_xy= h->mb_xy;
1482     const int mb_type= s->current_picture.mb_type[mb_xy];
1483     uint8_t  *dest_y, *dest_cb, *dest_cr;
1484     int linesize, uvlinesize /*dct_offset*/;
1485     int i;
1486     int *block_offset = &h->block_offset[0];
1487     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1488     /* is_h264 should always be true if SVQ3 is disabled. */
1489     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1490     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1491     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1492
1493     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1494     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1495     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1496
1497     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1498     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1499
1500     h->list_counts[mb_xy]= h->list_count;
1501
1502     if (!simple && MB_FIELD) {
1503         linesize   = h->mb_linesize   = s->linesize * 2;
1504         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1505         block_offset = &h->block_offset[24];
1506         if(mb_y&1){ //FIXME move out of this function?
1507             dest_y -= s->linesize*15;
1508             dest_cb-= s->uvlinesize*7;
1509             dest_cr-= s->uvlinesize*7;
1510         }
1511         if(FRAME_MBAFF) {
1512             int list;
1513             for(list=0; list<h->list_count; list++){
1514                 if(!USES_LIST(mb_type, list))
1515                     continue;
1516                 if(IS_16X16(mb_type)){
1517                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1518                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1519                 }else{
1520                     for(i=0; i<16; i+=4){
1521                         int ref = h->ref_cache[list][scan8[i]];
1522                         if(ref >= 0)
1523                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1524                     }
1525                 }
1526             }
1527         }
1528     } else {
1529         linesize   = h->mb_linesize   = s->linesize;
1530         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1531 //        dct_offset = s->linesize * 16;
1532     }
1533
1534     if (!simple && IS_INTRA_PCM(mb_type)) {
1535         for (i=0; i<16; i++) {
1536             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1537         }
1538         for (i=0; i<8; i++) {
1539             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1540             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1541         }
1542     } else {
1543         if(IS_INTRA(mb_type)){
1544             if(h->deblocking_filter)
1545                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1546
1547             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1548                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1549                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1550             }
1551
1552             if(IS_INTRA4x4(mb_type)){
1553                 if(simple || !s->encoding){
1554                     if(IS_8x8DCT(mb_type)){
1555                         if(transform_bypass){
1556                             idct_dc_add =
1557                             idct_add    = s->dsp.add_pixels8;
1558                         }else{
1559                             idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1560                             idct_add    = h->h264dsp.h264_idct8_add;
1561                         }
1562                         for(i=0; i<16; i+=4){
1563                             uint8_t * const ptr= dest_y + block_offset[i];
1564                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1565                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1566                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1567                             }else{
1568                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1569                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1570                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1571                                 if(nnz){
1572                                     if(nnz == 1 && h->mb[i*16])
1573                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1574                                     else
1575                                         idct_add   (ptr, h->mb + i*16, linesize);
1576                                 }
1577                             }
1578                         }
1579                     }else{
1580                         if(transform_bypass){
1581                             idct_dc_add =
1582                             idct_add    = s->dsp.add_pixels4;
1583                         }else{
1584                             idct_dc_add = h->h264dsp.h264_idct_dc_add;
1585                             idct_add    = h->h264dsp.h264_idct_add;
1586                         }
1587                         for(i=0; i<16; i++){
1588                             uint8_t * const ptr= dest_y + block_offset[i];
1589                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1590
1591                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1592                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1593                             }else{
1594                                 uint8_t *topright;
1595                                 int nnz, tr;
1596                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1597                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1598                                     assert(mb_y || linesize <= block_offset[i]);
1599                                     if(!topright_avail){
1600                                         tr= ptr[3 - linesize]*0x01010101;
1601                                         topright= (uint8_t*) &tr;
1602                                     }else
1603                                         topright= ptr + 4 - linesize;
1604                                 }else
1605                                     topright= NULL;
1606
1607                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1608                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1609                                 if(nnz){
1610                                     if(is_h264){
1611                                         if(nnz == 1 && h->mb[i*16])
1612                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1613                                         else
1614                                             idct_add   (ptr, h->mb + i*16, linesize);
1615                                     }else
1616                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1617                                 }
1618                             }
1619                         }
1620                     }
1621                 }
1622             }else{
1623                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1624                 if(is_h264){
1625                     if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX] ]){
1626                         if(!transform_bypass)
1627                             h->h264dsp.h264_luma_dc_dequant_idct(h->mb, h->mb_luma_dc, h->dequant4_coeff[0][s->qscale][0]);
1628                         else{
1629                             static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1630                                                                     8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1631                             for(i = 0; i < 16; i++)
1632                                 h->mb[dc_mapping[i]] = h->mb_luma_dc[i];
1633                         }
1634                     }
1635                 }else
1636                     ff_svq3_luma_dc_dequant_idct_c(h->mb, h->mb_luma_dc, s->qscale);
1637             }
1638             if(h->deblocking_filter)
1639                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1640         }else if(is_h264){
1641             hl_motion(h, dest_y, dest_cb, dest_cr,
1642                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1643                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1644                       h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
1645         }
1646
1647
1648         if(!IS_INTRA4x4(mb_type)){
1649             if(is_h264){
1650                 if(IS_INTRA16x16(mb_type)){
1651                     if(transform_bypass){
1652                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1653                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1654                         }else{
1655                             for(i=0; i<16; i++){
1656                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1657                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1658                             }
1659                         }
1660                     }else{
1661                          h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1662                     }
1663                 }else if(h->cbp&15){
1664                     if(transform_bypass){
1665                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1666                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1667                         for(i=0; i<16; i+=di){
1668                             if(h->non_zero_count_cache[ scan8[i] ]){
1669                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1670                             }
1671                         }
1672                     }else{
1673                         if(IS_8x8DCT(mb_type)){
1674                             h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1675                         }else{
1676                             h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1677                         }
1678                     }
1679                 }
1680             }else{
1681                 for(i=0; i<16; i++){
1682                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1683                         uint8_t * const ptr= dest_y + block_offset[i];
1684                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1685                     }
1686                 }
1687             }
1688         }
1689
1690         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1691             uint8_t *dest[2] = {dest_cb, dest_cr};
1692             if(transform_bypass){
1693                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1694                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1695                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1696                 }else{
1697                     idct_add = s->dsp.add_pixels4;
1698                     for(i=16; i<16+8; i++){
1699                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1700                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1701                     }
1702                 }
1703             }else{
1704                 if(is_h264){
1705                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
1706                         chroma_dc_dequant_idct_c(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1707                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
1708                         chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1709                     h->h264dsp.h264_idct_add8(dest, block_offset,
1710                                               h->mb, uvlinesize,
1711                                               h->non_zero_count_cache);
1712                 }else{
1713                     chroma_dc_dequant_idct_c(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1714                     chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1715                     for(i=16; i<16+8; i++){
1716                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1717                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1718                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1719                         }
1720                     }
1721                 }
1722             }
1723         }
1724     }
1725     if(h->cbp || IS_INTRA(mb_type))
1726         s->dsp.clear_blocks(h->mb);
1727 }
1728
1729 /**
1730  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1731  */
1732 static void hl_decode_mb_simple(H264Context *h){
1733     hl_decode_mb_internal(h, 1);
1734 }
1735
1736 /**
1737  * Process a macroblock; this handles edge cases, such as interlacing.
1738  */
1739 static void av_noinline hl_decode_mb_complex(H264Context *h){
1740     hl_decode_mb_internal(h, 0);
1741 }
1742
1743 void ff_h264_hl_decode_mb(H264Context *h){
1744     MpegEncContext * const s = &h->s;
1745     const int mb_xy= h->mb_xy;
1746     const int mb_type= s->current_picture.mb_type[mb_xy];
1747     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1748
1749     if (is_complex)
1750         hl_decode_mb_complex(h);
1751     else hl_decode_mb_simple(h);
1752 }
1753
1754 static int pred_weight_table(H264Context *h){
1755     MpegEncContext * const s = &h->s;
1756     int list, i;
1757     int luma_def, chroma_def;
1758
1759     h->use_weight= 0;
1760     h->use_weight_chroma= 0;
1761     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1762     if(CHROMA)
1763         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1764     luma_def = 1<<h->luma_log2_weight_denom;
1765     chroma_def = 1<<h->chroma_log2_weight_denom;
1766
1767     for(list=0; list<2; list++){
1768         h->luma_weight_flag[list]   = 0;
1769         h->chroma_weight_flag[list] = 0;
1770         for(i=0; i<h->ref_count[list]; i++){
1771             int luma_weight_flag, chroma_weight_flag;
1772
1773             luma_weight_flag= get_bits1(&s->gb);
1774             if(luma_weight_flag){
1775                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1776                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1777                 if(   h->luma_weight[i][list][0] != luma_def
1778                    || h->luma_weight[i][list][1] != 0) {
1779                     h->use_weight= 1;
1780                     h->luma_weight_flag[list]= 1;
1781                 }
1782             }else{
1783                 h->luma_weight[i][list][0]= luma_def;
1784                 h->luma_weight[i][list][1]= 0;
1785             }
1786
1787             if(CHROMA){
1788                 chroma_weight_flag= get_bits1(&s->gb);
1789                 if(chroma_weight_flag){
1790                     int j;
1791                     for(j=0; j<2; j++){
1792                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1793                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1794                         if(   h->chroma_weight[i][list][j][0] != chroma_def
1795                            || h->chroma_weight[i][list][j][1] != 0) {
1796                             h->use_weight_chroma= 1;
1797                             h->chroma_weight_flag[list]= 1;
1798                         }
1799                     }
1800                 }else{
1801                     int j;
1802                     for(j=0; j<2; j++){
1803                         h->chroma_weight[i][list][j][0]= chroma_def;
1804                         h->chroma_weight[i][list][j][1]= 0;
1805                     }
1806                 }
1807             }
1808         }
1809         if(h->slice_type_nos != FF_B_TYPE) break;
1810     }
1811     h->use_weight= h->use_weight || h->use_weight_chroma;
1812     return 0;
1813 }
1814
1815 /**
1816  * Initialize implicit_weight table.
1817  * @param field  0/1 initialize the weight for interlaced MBAFF
1818  *                -1 initializes the rest
1819  */
1820 static void implicit_weight_table(H264Context *h, int field){
1821     MpegEncContext * const s = &h->s;
1822     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1823
1824     for (i = 0; i < 2; i++) {
1825         h->luma_weight_flag[i]   = 0;
1826         h->chroma_weight_flag[i] = 0;
1827     }
1828
1829     if(field < 0){
1830         cur_poc = s->current_picture_ptr->poc;
1831     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1832        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1833         h->use_weight= 0;
1834         h->use_weight_chroma= 0;
1835         return;
1836     }
1837         ref_start= 0;
1838         ref_count0= h->ref_count[0];
1839         ref_count1= h->ref_count[1];
1840     }else{
1841         cur_poc = s->current_picture_ptr->field_poc[field];
1842         ref_start= 16;
1843         ref_count0= 16+2*h->ref_count[0];
1844         ref_count1= 16+2*h->ref_count[1];
1845     }
1846
1847     h->use_weight= 2;
1848     h->use_weight_chroma= 2;
1849     h->luma_log2_weight_denom= 5;
1850     h->chroma_log2_weight_denom= 5;
1851
1852     for(ref0=ref_start; ref0 < ref_count0; ref0++){
1853         int poc0 = h->ref_list[0][ref0].poc;
1854         for(ref1=ref_start; ref1 < ref_count1; ref1++){
1855             int poc1 = h->ref_list[1][ref1].poc;
1856             int td = av_clip(poc1 - poc0, -128, 127);
1857             int w= 32;
1858             if(td){
1859                 int tb = av_clip(cur_poc - poc0, -128, 127);
1860                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1861                 int dist_scale_factor = (tb*tx + 32) >> 8;
1862                 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1863                     w = 64 - dist_scale_factor;
1864             }
1865             if(field<0){
1866                 h->implicit_weight[ref0][ref1][0]=
1867                 h->implicit_weight[ref0][ref1][1]= w;
1868             }else{
1869                 h->implicit_weight[ref0][ref1][field]=w;
1870             }
1871         }
1872     }
1873 }
1874
1875 /**
1876  * instantaneous decoder refresh.
1877  */
1878 static void idr(H264Context *h){
1879     ff_h264_remove_all_refs(h);
1880     h->prev_frame_num= 0;
1881     h->prev_frame_num_offset= 0;
1882     h->prev_poc_msb=
1883     h->prev_poc_lsb= 0;
1884 }
1885
1886 /* forget old pics after a seek */
1887 static void flush_dpb(AVCodecContext *avctx){
1888     H264Context *h= avctx->priv_data;
1889     int i;
1890     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1891         if(h->delayed_pic[i])
1892             h->delayed_pic[i]->reference= 0;
1893         h->delayed_pic[i]= NULL;
1894     }
1895     h->outputed_poc=h->next_outputed_poc= INT_MIN;
1896     h->prev_interlaced_frame = 1;
1897     idr(h);
1898     if(h->s.current_picture_ptr)
1899         h->s.current_picture_ptr->reference= 0;
1900     h->s.first_field= 0;
1901     ff_h264_reset_sei(h);
1902     ff_mpeg_flush(avctx);
1903 }
1904
1905 static int init_poc(H264Context *h){
1906     MpegEncContext * const s = &h->s;
1907     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1908     int field_poc[2];
1909     Picture *cur = s->current_picture_ptr;
1910
1911     h->frame_num_offset= h->prev_frame_num_offset;
1912     if(h->frame_num < h->prev_frame_num)
1913         h->frame_num_offset += max_frame_num;
1914
1915     if(h->sps.poc_type==0){
1916         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1917
1918         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1919             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1920         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1921             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1922         else
1923             h->poc_msb = h->prev_poc_msb;
1924 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1925         field_poc[0] =
1926         field_poc[1] = h->poc_msb + h->poc_lsb;
1927         if(s->picture_structure == PICT_FRAME)
1928             field_poc[1] += h->delta_poc_bottom;
1929     }else if(h->sps.poc_type==1){
1930         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1931         int i;
1932
1933         if(h->sps.poc_cycle_length != 0)
1934             abs_frame_num = h->frame_num_offset + h->frame_num;
1935         else
1936             abs_frame_num = 0;
1937
1938         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1939             abs_frame_num--;
1940
1941         expected_delta_per_poc_cycle = 0;
1942         for(i=0; i < h->sps.poc_cycle_length; i++)
1943             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1944
1945         if(abs_frame_num > 0){
1946             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1947             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1948
1949             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1950             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1951                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1952         } else
1953             expectedpoc = 0;
1954
1955         if(h->nal_ref_idc == 0)
1956             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1957
1958         field_poc[0] = expectedpoc + h->delta_poc[0];
1959         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1960
1961         if(s->picture_structure == PICT_FRAME)
1962             field_poc[1] += h->delta_poc[1];
1963     }else{
1964         int poc= 2*(h->frame_num_offset + h->frame_num);
1965
1966         if(!h->nal_ref_idc)
1967             poc--;
1968
1969         field_poc[0]= poc;
1970         field_poc[1]= poc;
1971     }
1972
1973     if(s->picture_structure != PICT_BOTTOM_FIELD)
1974         s->current_picture_ptr->field_poc[0]= field_poc[0];
1975     if(s->picture_structure != PICT_TOP_FIELD)
1976         s->current_picture_ptr->field_poc[1]= field_poc[1];
1977     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1978
1979     return 0;
1980 }
1981
1982
1983 /**
1984  * initialize scan tables
1985  */
1986 static void init_scan_tables(H264Context *h){
1987     int i;
1988     for(i=0; i<16; i++){
1989 #define T(x) (x>>2) | ((x<<2) & 0xF)
1990         h->zigzag_scan[i] = T(zigzag_scan[i]);
1991         h-> field_scan[i] = T( field_scan[i]);
1992 #undef T
1993     }
1994     for(i=0; i<64; i++){
1995 #define T(x) (x>>3) | ((x&7)<<3)
1996         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1997         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1998         h->field_scan8x8[i]        = T(field_scan8x8[i]);
1999         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
2000 #undef T
2001     }
2002     if(h->sps.transform_bypass){ //FIXME same ugly
2003         h->zigzag_scan_q0          = zigzag_scan;
2004         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
2005         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2006         h->field_scan_q0           = field_scan;
2007         h->field_scan8x8_q0        = field_scan8x8;
2008         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
2009     }else{
2010         h->zigzag_scan_q0          = h->zigzag_scan;
2011         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
2012         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2013         h->field_scan_q0           = h->field_scan;
2014         h->field_scan8x8_q0        = h->field_scan8x8;
2015         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
2016     }
2017 }
2018
2019 static void field_end(H264Context *h, int in_setup){
2020     MpegEncContext * const s = &h->s;
2021     AVCodecContext * const avctx= s->avctx;
2022     s->mb_y= 0;
2023
2024     if (!in_setup && !s->dropable)
2025         ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
2026                                  s->picture_structure==PICT_BOTTOM_FIELD);
2027
2028     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2029         ff_vdpau_h264_set_reference_frames(s);
2030
2031     if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2032         if(!s->dropable) {
2033             ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2034             h->prev_poc_msb= h->poc_msb;
2035             h->prev_poc_lsb= h->poc_lsb;
2036         }
2037         h->prev_frame_num_offset= h->frame_num_offset;
2038         h->prev_frame_num= h->frame_num;
2039         h->outputed_poc = h->next_outputed_poc;
2040     }
2041
2042     if (avctx->hwaccel) {
2043         if (avctx->hwaccel->end_frame(avctx) < 0)
2044             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2045     }
2046
2047     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2048         ff_vdpau_h264_picture_complete(s);
2049
2050     /*
2051      * FIXME: Error handling code does not seem to support interlaced
2052      * when slices span multiple rows
2053      * The ff_er_add_slice calls don't work right for bottom
2054      * fields; they cause massive erroneous error concealing
2055      * Error marking covers both fields (top and bottom).
2056      * This causes a mismatched s->error_count
2057      * and a bad error table. Further, the error count goes to
2058      * INT_MAX when called for bottom field, because mb_y is
2059      * past end by one (callers fault) and resync_mb_y != 0
2060      * causes problems for the first MB line, too.
2061      */
2062     if (!FIELD_PICTURE)
2063         ff_er_frame_end(s);
2064
2065     MPV_frame_end(s);
2066
2067     h->current_slice=0;
2068 }
2069
2070 /**
2071  * Replicate H264 "master" context to thread contexts.
2072  */
2073 static void clone_slice(H264Context *dst, H264Context *src)
2074 {
2075     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
2076     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
2077     dst->s.current_picture      = src->s.current_picture;
2078     dst->s.linesize             = src->s.linesize;
2079     dst->s.uvlinesize           = src->s.uvlinesize;
2080     dst->s.first_field          = src->s.first_field;
2081
2082     dst->prev_poc_msb           = src->prev_poc_msb;
2083     dst->prev_poc_lsb           = src->prev_poc_lsb;
2084     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
2085     dst->prev_frame_num         = src->prev_frame_num;
2086     dst->short_ref_count        = src->short_ref_count;
2087
2088     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
2089     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
2090     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2091     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
2092
2093     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
2094     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
2095 }
2096
2097 /**
2098  * computes profile from profile_idc and constraint_set?_flags
2099  *
2100  * @param sps SPS
2101  *
2102  * @return profile as defined by FF_PROFILE_H264_*
2103  */
2104 int ff_h264_get_profile(SPS *sps)
2105 {
2106     int profile = sps->profile_idc;
2107
2108     switch(sps->profile_idc) {
2109     case FF_PROFILE_H264_BASELINE:
2110         // constraint_set1_flag set to 1
2111         profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2112         break;
2113     case FF_PROFILE_H264_HIGH_10:
2114     case FF_PROFILE_H264_HIGH_422:
2115     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2116         // constraint_set3_flag set to 1
2117         profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2118         break;
2119     }
2120
2121     return profile;
2122 }
2123
2124 /**
2125  * decodes a slice header.
2126  * This will also call MPV_common_init() and frame_start() as needed.
2127  *
2128  * @param h h264context
2129  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2130  *
2131  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2132  */
2133 static int decode_slice_header(H264Context *h, H264Context *h0){
2134     MpegEncContext * const s = &h->s;
2135     MpegEncContext * const s0 = &h0->s;
2136     unsigned int first_mb_in_slice;
2137     unsigned int pps_id;
2138     int num_ref_idx_active_override_flag;
2139     unsigned int slice_type, tmp, i, j;
2140     int default_ref_list_done = 0;
2141     int last_pic_structure;
2142
2143     s->dropable= h->nal_ref_idc == 0;
2144
2145     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
2146         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2147         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2148     }else{
2149         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2150         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2151     }
2152
2153     first_mb_in_slice= get_ue_golomb(&s->gb);
2154
2155     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2156         if(h0->current_slice && FIELD_PICTURE){
2157             field_end(h, 1);
2158         }
2159
2160         h0->current_slice = 0;
2161         if (!s0->first_field)
2162             s->current_picture_ptr= NULL;
2163     }
2164
2165     slice_type= get_ue_golomb_31(&s->gb);
2166     if(slice_type > 9){
2167         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);
2168         return -1;
2169     }
2170     if(slice_type > 4){
2171         slice_type -= 5;
2172         h->slice_type_fixed=1;
2173     }else
2174         h->slice_type_fixed=0;
2175
2176     slice_type= golomb_to_pict_type[ slice_type ];
2177     if (slice_type == FF_I_TYPE
2178         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
2179         default_ref_list_done = 1;
2180     }
2181     h->slice_type= slice_type;
2182     h->slice_type_nos= slice_type & 3;
2183
2184     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
2185
2186     pps_id= get_ue_golomb(&s->gb);
2187     if(pps_id>=MAX_PPS_COUNT){
2188         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2189         return -1;
2190     }
2191     if(!h0->pps_buffers[pps_id]) {
2192         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
2193         return -1;
2194     }
2195     h->pps= *h0->pps_buffers[pps_id];
2196
2197     if(!h0->sps_buffers[h->pps.sps_id]) {
2198         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
2199         return -1;
2200     }
2201     h->sps = *h0->sps_buffers[h->pps.sps_id];
2202
2203     s->avctx->profile = ff_h264_get_profile(&h->sps);
2204     s->avctx->level   = h->sps.level_idc;
2205     s->avctx->refs    = h->sps.ref_frame_count;
2206
2207     if(h == h0 && h->dequant_coeff_pps != pps_id){
2208         h->dequant_coeff_pps = pps_id;
2209         init_dequant_tables(h);
2210     }
2211
2212     s->mb_width= h->sps.mb_width;
2213     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2214
2215     h->b_stride=  s->mb_width*4;
2216
2217     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
2218     if(h->sps.frame_mbs_only_flag)
2219         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
2220     else
2221         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 7);
2222
2223     if (s->context_initialized
2224         && (   s->width != s->avctx->width || s->height != s->avctx->height
2225             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
2226         if(h != h0) {
2227             av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
2228             return -1;   // width / height changed during parallelized decoding
2229         }
2230         free_tables(h, 0);
2231         flush_dpb(s->avctx);
2232         MPV_common_end(s);
2233     }
2234     if (!s->context_initialized) {
2235         if(h != h0)
2236             return -1;  // we cant (re-)initialize context during parallel decoding
2237
2238         avcodec_set_dimensions(s->avctx, s->width, s->height);
2239         s->avctx->sample_aspect_ratio= h->sps.sar;
2240         av_assert0(s->avctx->sample_aspect_ratio.den);
2241
2242         if(h->sps.video_signal_type_present_flag){
2243             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2244             if(h->sps.colour_description_present_flag){
2245                 s->avctx->color_primaries = h->sps.color_primaries;
2246                 s->avctx->color_trc       = h->sps.color_trc;
2247                 s->avctx->colorspace      = h->sps.colorspace;
2248             }
2249         }
2250
2251         if(h->sps.timing_info_present_flag){
2252             int64_t den= h->sps.time_scale;
2253             if(h->x264_build < 44U)
2254                 den *= 2;
2255             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2256                       h->sps.num_units_in_tick, den, 1<<30);
2257         }
2258         s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2259                                                  s->avctx->codec->pix_fmts ?
2260                                                  s->avctx->codec->pix_fmts :
2261                                                  s->avctx->color_range == AVCOL_RANGE_JPEG ?
2262                                                  hwaccel_pixfmt_list_h264_jpeg_420 :
2263                                                  ff_hwaccel_pixfmt_list_420);
2264         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2265
2266         if (MPV_common_init(s) < 0)
2267             return -1;
2268         s->first_field = 0;
2269         h->prev_interlaced_frame = 1;
2270
2271         init_scan_tables(h);
2272         ff_h264_alloc_tables(h);
2273
2274         if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
2275             if (context_init(h) < 0)
2276                 return -1;
2277         } else {
2278             for(i = 1; i < s->avctx->thread_count; i++) {
2279                 H264Context *c;
2280                 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2281                 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2282                 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2283                 c->h264dsp = h->h264dsp;
2284                 c->sps = h->sps;
2285                 c->pps = h->pps;
2286                 init_scan_tables(c);
2287                 clone_tables(c, h, i);
2288             }
2289
2290             for(i = 0; i < s->avctx->thread_count; i++)
2291                 if(context_init(h->thread_context[i]) < 0)
2292                     return -1;
2293         }
2294     }
2295
2296     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2297
2298     h->mb_mbaff = 0;
2299     h->mb_aff_frame = 0;
2300     last_pic_structure = s0->picture_structure;
2301     if(h->sps.frame_mbs_only_flag){
2302         s->picture_structure= PICT_FRAME;
2303     }else{
2304         if(get_bits1(&s->gb)) { //field_pic_flag
2305             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2306         } else {
2307             s->picture_structure= PICT_FRAME;
2308             h->mb_aff_frame = h->sps.mb_aff;
2309         }
2310     }
2311     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2312
2313     if(h0->current_slice == 0){
2314         if(h->frame_num != h->prev_frame_num &&
2315           (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num) < (h->frame_num - h->sps.ref_frame_count))
2316             h->prev_frame_num = h->frame_num - h->sps.ref_frame_count - 1;
2317
2318         while(h->frame_num !=  h->prev_frame_num &&
2319               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
2320             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2321             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
2322             if (ff_h264_frame_start(h) < 0)
2323                 return -1;
2324             h->prev_frame_num++;
2325             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2326             s->current_picture_ptr->frame_num= h->prev_frame_num;
2327             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
2328             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
2329             ff_generate_sliding_window_mmcos(h);
2330             ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2331             /* Error concealment: if a ref is missing, copy the previous ref in its place.
2332              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2333              * about there being no actual duplicates.
2334              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
2335              * concealing a lost frame, this probably isn't noticable by comparison, but it should
2336              * be fixed. */
2337             if (h->short_ref_count) {
2338                 if (prev) {
2339                     av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
2340                                   (const uint8_t**)prev->data, prev->linesize,
2341                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2342                     h->short_ref[0]->poc = prev->poc+2;
2343                 }
2344                 h->short_ref[0]->frame_num = h->prev_frame_num;
2345             }
2346         }
2347
2348         /* See if we have a decoded first field looking for a pair... */
2349         if (s0->first_field) {
2350             assert(s0->current_picture_ptr);
2351             assert(s0->current_picture_ptr->data[0]);
2352             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
2353
2354             /* figure out if we have a complementary field pair */
2355             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2356                 /*
2357                  * Previous field is unmatched. Don't display it, but let it
2358                  * remain for reference if marked as such.
2359                  */
2360                 s0->current_picture_ptr = NULL;
2361                 s0->first_field = FIELD_PICTURE;
2362
2363             } else {
2364                 if (h->nal_ref_idc &&
2365                         s0->current_picture_ptr->reference &&
2366                         s0->current_picture_ptr->frame_num != h->frame_num) {
2367                     /*
2368                      * This and previous field were reference, but had
2369                      * different frame_nums. Consider this field first in
2370                      * pair. Throw away previous field except for reference
2371                      * purposes.
2372                      */
2373                     s0->first_field = 1;
2374                     s0->current_picture_ptr = NULL;
2375
2376                 } else {
2377                     /* Second field in complementary pair */
2378                     s0->first_field = 0;
2379                 }
2380             }
2381
2382         } else {
2383             /* Frame or first field in a potentially complementary pair */
2384             assert(!s0->current_picture_ptr);
2385             s0->first_field = FIELD_PICTURE;
2386         }
2387
2388         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
2389             s0->first_field = 0;
2390             return -1;
2391         }
2392     }
2393     if(h != h0)
2394         clone_slice(h, h0);
2395
2396     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2397
2398     assert(s->mb_num == s->mb_width * s->mb_height);
2399     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2400        first_mb_in_slice                    >= s->mb_num){
2401         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2402         return -1;
2403     }
2404     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2405     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2406     if (s->picture_structure == PICT_BOTTOM_FIELD)
2407         s->resync_mb_y = s->mb_y = s->mb_y + 1;
2408     assert(s->mb_y < s->mb_height);
2409
2410     if(s->picture_structure==PICT_FRAME){
2411         h->curr_pic_num=   h->frame_num;
2412         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2413     }else{
2414         h->curr_pic_num= 2*h->frame_num + 1;
2415         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2416     }
2417
2418     if(h->nal_unit_type == NAL_IDR_SLICE){
2419         get_ue_golomb(&s->gb); /* idr_pic_id */
2420     }
2421
2422     if(h->sps.poc_type==0){
2423         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2424
2425         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2426             h->delta_poc_bottom= get_se_golomb(&s->gb);
2427         }
2428     }
2429
2430     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2431         h->delta_poc[0]= get_se_golomb(&s->gb);
2432
2433         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2434             h->delta_poc[1]= get_se_golomb(&s->gb);
2435     }
2436
2437     init_poc(h);
2438
2439     if(h->pps.redundant_pic_cnt_present){
2440         h->redundant_pic_count= get_ue_golomb(&s->gb);
2441     }
2442
2443     //set defaults, might be overridden a few lines later
2444     h->ref_count[0]= h->pps.ref_count[0];
2445     h->ref_count[1]= h->pps.ref_count[1];
2446
2447     if(h->slice_type_nos != FF_I_TYPE){
2448         if(h->slice_type_nos == FF_B_TYPE){
2449             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2450         }
2451         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2452
2453         if(num_ref_idx_active_override_flag){
2454             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2455             if(h->slice_type_nos==FF_B_TYPE)
2456                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2457
2458             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2459                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2460                 h->ref_count[0]= h->ref_count[1]= 1;
2461                 return -1;
2462             }
2463         }
2464         if(h->slice_type_nos == FF_B_TYPE)
2465             h->list_count= 2;
2466         else
2467             h->list_count= 1;
2468     }else
2469         h->list_count= 0;
2470
2471     if(!default_ref_list_done){
2472         ff_h264_fill_default_ref_list(h);
2473     }
2474
2475     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2476         return -1;
2477
2478     //FIXME mt gives valgrind warnings and crashes if this is uncommented
2479     /*
2480
2481     if(h->slice_type_nos!=FF_I_TYPE){
2482         s->last_picture_ptr= &h->ref_list[0][0];
2483         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2484     }
2485     if(h->slice_type_nos==FF_B_TYPE){
2486         s->next_picture_ptr= &h->ref_list[1][0];
2487         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2488     }
2489
2490     */
2491
2492     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2493        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2494         pred_weight_table(h);
2495     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2496         implicit_weight_table(h, -1);
2497     }else {
2498         h->use_weight = 0;
2499         for (i = 0; i < 2; i++) {
2500             h->luma_weight_flag[i]   = 0;
2501             h->chroma_weight_flag[i] = 0;
2502         }
2503     }
2504
2505     if(h->nal_ref_idc)
2506         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2507
2508     if(FRAME_MBAFF){
2509         ff_h264_fill_mbaff_ref_list(h);
2510
2511         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2512             implicit_weight_table(h, 0);
2513             implicit_weight_table(h, 1);
2514         }
2515     }
2516
2517     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2518         ff_h264_direct_dist_scale_factor(h);
2519     ff_h264_direct_ref_list_init(h);
2520
2521     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2522         tmp = get_ue_golomb_31(&s->gb);
2523         if(tmp > 2){
2524             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2525             return -1;
2526         }
2527         h->cabac_init_idc= tmp;
2528     }
2529
2530     h->last_qscale_diff = 0;
2531     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2532     if(tmp>51){
2533         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2534         return -1;
2535     }
2536     s->qscale= tmp;
2537     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2538     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2539     //FIXME qscale / qp ... stuff
2540     if(h->slice_type == FF_SP_TYPE){
2541         get_bits1(&s->gb); /* sp_for_switch_flag */
2542     }
2543     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2544         get_se_golomb(&s->gb); /* slice_qs_delta */
2545     }
2546
2547     h->deblocking_filter = 1;
2548     h->slice_alpha_c0_offset = 52;
2549     h->slice_beta_offset = 52;
2550     if( h->pps.deblocking_filter_parameters_present ) {
2551         tmp= get_ue_golomb_31(&s->gb);
2552         if(tmp > 2){
2553             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2554             return -1;
2555         }
2556         h->deblocking_filter= tmp;
2557         if(h->deblocking_filter < 2)
2558             h->deblocking_filter^= 1; // 1<->0
2559
2560         if( h->deblocking_filter ) {
2561             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2562             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2563             if(   h->slice_alpha_c0_offset > 104U
2564                || h->slice_beta_offset     > 104U){
2565                 av_log(s->avctx, AV_LOG_ERROR, "deblocking filter parameters %d %d out of range\n", h->slice_alpha_c0_offset, h->slice_beta_offset);
2566                 return -1;
2567             }
2568         }
2569     }
2570
2571     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2572        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2573        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2574        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2575         h->deblocking_filter= 0;
2576
2577     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2578         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2579             /* Cheat slightly for speed:
2580                Do not bother to deblock across slices. */
2581             h->deblocking_filter = 2;
2582         } else {
2583             h0->max_contexts = 1;
2584             if(!h0->single_decode_warning) {
2585                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2586                 h0->single_decode_warning = 1;
2587             }
2588             if(h != h0)
2589                 return 1; // deblocking switched inside frame
2590         }
2591     }
2592     h->qp_thresh= 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
2593
2594 #if 0 //FMO
2595     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2596         slice_group_change_cycle= get_bits(&s->gb, ?);
2597 #endif
2598
2599     h0->last_slice_type = slice_type;
2600     h->slice_num = ++h0->current_slice;
2601     if(h->slice_num >= MAX_SLICES){
2602         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2603     }
2604
2605     for(j=0; j<2; j++){
2606         int id_list[16];
2607         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2608         for(i=0; i<16; i++){
2609             id_list[i]= 60;
2610             if(h->ref_list[j][i].data[0]){
2611                 int k;
2612                 uint8_t *base= h->ref_list[j][i].base[0];
2613                 for(k=0; k<h->short_ref_count; k++)
2614                     if(h->short_ref[k]->base[0] == base){
2615                         id_list[i]= k;
2616                         break;
2617                     }
2618                 for(k=0; k<h->long_ref_count; k++)
2619                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2620                         id_list[i]= h->short_ref_count + k;
2621                         break;
2622                     }
2623             }
2624         }
2625
2626         ref2frm[0]=
2627         ref2frm[1]= -1;
2628         for(i=0; i<16; i++)
2629             ref2frm[i+2]= 4*id_list[i]
2630                           +(h->ref_list[j][i].reference&3);
2631         ref2frm[18+0]=
2632         ref2frm[18+1]= -1;
2633         for(i=16; i<48; i++)
2634             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2635                           +(h->ref_list[j][i].reference&3);
2636     }
2637
2638     //FIXME: fix draw_edges+PAFF+frame threads
2639     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE || (!h->sps.frame_mbs_only_flag && s->avctx->active_thread_type&FF_THREAD_FRAME)) ? 0 : 16;
2640     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2641
2642     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2643         av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
2644                h->slice_num,
2645                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2646                first_mb_in_slice,
2647                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2648                pps_id, h->frame_num,
2649                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2650                h->ref_count[0], h->ref_count[1],
2651                s->qscale,
2652                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2653                h->use_weight,
2654                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2655                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2656                );
2657     }
2658
2659     return 0;
2660 }
2661
2662 int ff_h264_get_slice_type(const H264Context *h)
2663 {
2664     switch (h->slice_type) {
2665     case FF_P_TYPE:  return 0;
2666     case FF_B_TYPE:  return 1;
2667     case FF_I_TYPE:  return 2;
2668     case FF_SP_TYPE: return 3;
2669     case FF_SI_TYPE: return 4;
2670     default:         return -1;
2671     }
2672 }
2673
2674 /**
2675  *
2676  * @return non zero if the loop filter can be skiped
2677  */
2678 static int fill_filter_caches(H264Context *h, int mb_type){
2679     MpegEncContext * const s = &h->s;
2680     const int mb_xy= h->mb_xy;
2681     int top_xy, left_xy[2];
2682     int top_type, left_type[2];
2683
2684     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
2685
2686     //FIXME deblocking could skip the intra and nnz parts.
2687
2688     /* Wow, what a mess, why didn't they simplify the interlacing & intra
2689      * stuff, I can't imagine that these complex rules are worth it. */
2690
2691     left_xy[1] = left_xy[0] = mb_xy-1;
2692     if(FRAME_MBAFF){
2693         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2694         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
2695         if(s->mb_y&1){
2696             if (left_mb_field_flag != curr_mb_field_flag) {
2697                 left_xy[0] -= s->mb_stride;
2698             }
2699         }else{
2700             if(curr_mb_field_flag){
2701                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
2702             }
2703             if (left_mb_field_flag != curr_mb_field_flag) {
2704                 left_xy[1] += s->mb_stride;
2705             }
2706         }
2707     }
2708
2709     h->top_mb_xy = top_xy;
2710     h->left_mb_xy[0] = left_xy[0];
2711     h->left_mb_xy[1] = left_xy[1];
2712     {
2713         //for sufficiently low qp, filtering wouldn't do anything
2714         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2715         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2716         int qp = s->current_picture.qscale_table[mb_xy];
2717         if(qp <= qp_thresh
2718            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2719            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
2720             if(!FRAME_MBAFF)
2721                 return 1;
2722             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
2723                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
2724                 return 1;
2725         }
2726     }
2727
2728     top_type     = s->current_picture.mb_type[top_xy]    ;
2729     left_type[0] = s->current_picture.mb_type[left_xy[0]];
2730     left_type[1] = s->current_picture.mb_type[left_xy[1]];
2731     if(h->deblocking_filter == 2){
2732         if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
2733         if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2734     }else{
2735         if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
2736         if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2737     }
2738     h->top_type    = top_type    ;
2739     h->left_type[0]= left_type[0];
2740     h->left_type[1]= left_type[1];
2741
2742     if(IS_INTRA(mb_type))
2743         return 0;
2744
2745     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2746     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2747     AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2748     AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2749     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2750
2751     h->cbp= h->cbp_table[mb_xy];
2752
2753     {
2754         int list;
2755         for(list=0; list<h->list_count; list++){
2756             int8_t *ref;
2757             int y, b_stride;
2758             int16_t (*mv_dst)[2];
2759             int16_t (*mv_src)[2];
2760
2761             if(!USES_LIST(mb_type, list)){
2762                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2763                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2764                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2765                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2766                 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2767                 continue;
2768             }
2769
2770             ref = &s->current_picture.ref_index[list][4*mb_xy];
2771             {
2772                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2773                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2774                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2775                 ref += 2;
2776                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2777                 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2778             }
2779
2780             b_stride = h->b_stride;
2781             mv_dst   = &h->mv_cache[list][scan8[0]];
2782             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2783             for(y=0; y<4; y++){
2784                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2785             }
2786
2787         }
2788     }
2789
2790
2791 /*
2792 0 . T T. T T T T
2793 1 L . .L . . . .
2794 2 L . .L . . . .
2795 3 . T TL . . . .
2796 4 L . .L . . . .
2797 5 L . .. . . . .
2798 */
2799 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2800     if(top_type){
2801         AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2802     }
2803
2804     if(left_type[0]){
2805         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2806         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2807         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2808         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2809     }
2810
2811     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2812     if(!CABAC && h->pps.transform_8x8_mode){
2813         if(IS_8x8DCT(top_type)){
2814             h->non_zero_count_cache[4+8*0]=
2815             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2816             h->non_zero_count_cache[6+8*0]=
2817             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2818         }
2819         if(IS_8x8DCT(left_type[0])){
2820             h->non_zero_count_cache[3+8*1]=
2821             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2822         }
2823         if(IS_8x8DCT(left_type[1])){
2824             h->non_zero_count_cache[3+8*3]=
2825             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2826         }
2827
2828         if(IS_8x8DCT(mb_type)){
2829             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
2830             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
2831
2832             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2833             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2834
2835             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2836             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2837
2838             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2839             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2840         }
2841     }
2842
2843     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2844         int list;
2845         for(list=0; list<h->list_count; list++){
2846             if(USES_LIST(top_type, list)){
2847                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2848                 const int b8_xy= 4*top_xy + 2;
2849                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2850                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2851                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2852                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2853                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2854                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2855             }else{
2856                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2857                 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2858             }
2859
2860             if(!IS_INTERLACED(mb_type^left_type[0])){
2861                 if(USES_LIST(left_type[0], list)){
2862                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2863                     const int b8_xy= 4*left_xy[0] + 1;
2864                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2865                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2866                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2867                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2868                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2869                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
2870                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2871                     h->ref_cache[list][scan8[0] - 1 +16 ]=
2872                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2873                 }else{
2874                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2875                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2876                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2877                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2878                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
2879                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
2880                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
2881                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2882                 }
2883             }
2884         }
2885     }
2886
2887     return 0;
2888 }
2889
2890 static void loop_filter(H264Context *h){
2891     MpegEncContext * const s = &h->s;
2892     uint8_t  *dest_y, *dest_cb, *dest_cr;
2893     int linesize, uvlinesize, mb_x, mb_y;
2894     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2895     const int old_slice_type= h->slice_type;
2896
2897     if(h->deblocking_filter) {
2898         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2899             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2900                 int mb_xy, mb_type;
2901                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2902                 h->slice_num= h->slice_table[mb_xy];
2903                 mb_type= s->current_picture.mb_type[mb_xy];
2904                 h->list_count= h->list_counts[mb_xy];
2905
2906                 if(FRAME_MBAFF)
2907                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2908
2909                 s->mb_x= mb_x;
2910                 s->mb_y= mb_y;
2911                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2912                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2913                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2914                     //FIXME simplify above
2915
2916                 if (MB_FIELD) {
2917                     linesize   = h->mb_linesize   = s->linesize * 2;
2918                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2919                     if(mb_y&1){ //FIXME move out of this function?
2920                         dest_y -= s->linesize*15;
2921                         dest_cb-= s->uvlinesize*7;
2922                         dest_cr-= s->uvlinesize*7;
2923                     }
2924                 } else {
2925                     linesize   = h->mb_linesize   = s->linesize;
2926                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2927                 }
2928                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2929                 if(fill_filter_caches(h, mb_type))
2930                     continue;
2931                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2932                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2933
2934                 if (FRAME_MBAFF) {
2935                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2936                 } else {
2937                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2938                 }
2939             }
2940         }
2941     }
2942     h->slice_type= old_slice_type;
2943     s->mb_x= 0;
2944     s->mb_y= end_mb_y - FRAME_MBAFF;
2945     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2946     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2947 }
2948
2949 static void predict_field_decoding_flag(H264Context *h){
2950     MpegEncContext * const s = &h->s;
2951     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2952     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2953                 ? s->current_picture.mb_type[mb_xy-1]
2954                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2955                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2956                 : 0;
2957     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2958 }
2959
2960 /**
2961  * Draw edges and report progress for the last MB row.
2962  */
2963 static void decode_finish_row(H264Context *h){
2964     MpegEncContext * const s = &h->s;
2965     int top = 16*(s->mb_y >> FIELD_PICTURE);
2966     int height = 16 << FRAME_MBAFF;
2967     int deblock_border = (16 + 4) << FRAME_MBAFF;
2968     int pic_height = 16*s->mb_height >> FIELD_PICTURE;
2969
2970     if (h->deblocking_filter) {
2971         if((top + height) >= pic_height)
2972             height += deblock_border;
2973
2974         top -= deblock_border;
2975     }
2976
2977     if (top >= pic_height || (top + height) < h->emu_edge_height)
2978         return;
2979
2980     height = FFMIN(height, pic_height - top);
2981     if (top < h->emu_edge_height) {
2982         height = top+height;
2983         top = 0;
2984     }
2985
2986     ff_draw_horiz_band(s, top, height);
2987
2988     if (s->dropable) return;
2989
2990     ff_thread_report_progress((AVFrame*)s->current_picture_ptr, top + height - 1,
2991                              s->picture_structure==PICT_BOTTOM_FIELD);
2992 }
2993
2994 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2995     H264Context *h = *(void**)arg;
2996     MpegEncContext * const s = &h->s;
2997     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2998
2999     s->mb_skip_run= -1;
3000
3001     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
3002                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
3003
3004     if( h->pps.cabac ) {
3005         /* realign */
3006         align_get_bits( &s->gb );
3007
3008         /* init cabac */
3009         ff_init_cabac_states( &h->cabac);
3010         ff_init_cabac_decoder( &h->cabac,
3011                                s->gb.buffer + get_bits_count(&s->gb)/8,
3012                                (get_bits_left(&s->gb) + 7)/8);
3013
3014         ff_h264_init_cabac_states(h);
3015
3016         for(;;){
3017 //START_TIMER
3018             int ret = ff_h264_decode_mb_cabac(h);
3019             int eos;
3020 //STOP_TIMER("decode_mb_cabac")
3021
3022             if(ret>=0) ff_h264_hl_decode_mb(h);
3023
3024             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
3025                 s->mb_y++;
3026
3027                 ret = ff_h264_decode_mb_cabac(h);
3028
3029                 if(ret>=0) ff_h264_hl_decode_mb(h);
3030                 s->mb_y--;
3031             }
3032             eos = get_cabac_terminate( &h->cabac );
3033
3034             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
3035                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3036                 return 0;
3037             }
3038             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3039                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%td)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);
3040                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3041                 return -1;
3042             }
3043
3044             if( ++s->mb_x >= s->mb_width ) {
3045                 s->mb_x = 0;
3046                 loop_filter(h);
3047                 decode_finish_row(h);
3048                 ++s->mb_y;
3049                 if(FIELD_OR_MBAFF_PICTURE) {
3050                     ++s->mb_y;
3051                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
3052                         predict_field_decoding_flag(h);
3053                 }
3054             }
3055
3056             if( eos || s->mb_y >= s->mb_height ) {
3057                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3058                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3059                 return 0;
3060             }
3061         }
3062
3063     } else {
3064         for(;;){
3065             int ret = ff_h264_decode_mb_cavlc(h);
3066
3067             if(ret>=0) ff_h264_hl_decode_mb(h);
3068
3069             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
3070                 s->mb_y++;
3071                 ret = ff_h264_decode_mb_cavlc(h);
3072
3073                 if(ret>=0) ff_h264_hl_decode_mb(h);
3074                 s->mb_y--;
3075             }
3076
3077             if(ret<0){
3078                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3079                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3080
3081                 return -1;
3082             }
3083
3084             if(++s->mb_x >= s->mb_width){
3085                 s->mb_x=0;
3086                 loop_filter(h);
3087                 decode_finish_row(h);
3088                 ++s->mb_y;
3089                 if(FIELD_OR_MBAFF_PICTURE) {
3090                     ++s->mb_y;
3091                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
3092                         predict_field_decoding_flag(h);
3093                 }
3094                 if(s->mb_y >= s->mb_height){
3095                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3096
3097                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
3098                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3099
3100                         return 0;
3101                     }else{
3102                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3103
3104                         return -1;
3105                     }
3106                 }
3107             }
3108
3109             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3110                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3111                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
3112                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3113
3114                     return 0;
3115                 }else{
3116                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3117
3118                     return -1;
3119                 }
3120             }
3121         }
3122     }
3123
3124 #if 0
3125     for(;s->mb_y < s->mb_height; s->mb_y++){
3126         for(;s->mb_x < s->mb_width; s->mb_x++){
3127             int ret= decode_mb(h);
3128
3129             ff_h264_hl_decode_mb(h);
3130
3131             if(ret<0){
3132                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3133                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3134
3135                 return -1;
3136             }
3137
3138             if(++s->mb_x >= s->mb_width){
3139                 s->mb_x=0;
3140                 if(++s->mb_y >= s->mb_height){
3141                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
3142                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3143
3144                         return 0;
3145                     }else{
3146                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3147
3148                         return -1;
3149                     }
3150                 }
3151             }
3152
3153             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
3154                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
3155                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3156
3157                     return 0;
3158                 }else{
3159                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3160
3161                     return -1;
3162                 }
3163             }
3164         }
3165         s->mb_x=0;
3166         ff_draw_horiz_band(s, 16*s->mb_y, 16);
3167     }
3168 #endif
3169     return -1; //not reached
3170 }
3171
3172 /**
3173  * Call decode_slice() for each context.
3174  *
3175  * @param h h264 master context
3176  * @param context_count number of contexts to execute
3177  */
3178 static void execute_decode_slices(H264Context *h, int context_count){
3179     MpegEncContext * const s = &h->s;
3180     AVCodecContext * const avctx= s->avctx;
3181     H264Context *hx;
3182     int i;
3183
3184     if (s->avctx->hwaccel)
3185         return;
3186     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3187         return;
3188     if(context_count == 1) {
3189         decode_slice(avctx, &h);
3190     } else {
3191         for(i = 1; i < context_count; i++) {
3192             hx = h->thread_context[i];
3193             hx->s.error_recognition = avctx->error_recognition;
3194             hx->s.error_count = 0;
3195         }
3196
3197         avctx->execute(avctx, (void *)decode_slice,
3198                        h->thread_context, NULL, context_count, sizeof(void*));
3199
3200         /* pull back stuff from slices to master context */
3201         hx = h->thread_context[context_count - 1];
3202         s->mb_x = hx->s.mb_x;
3203         s->mb_y = hx->s.mb_y;
3204         s->dropable = hx->s.dropable;
3205         s->picture_structure = hx->s.picture_structure;
3206         for(i = 1; i < context_count; i++)
3207             h->s.error_count += h->thread_context[i]->s.error_count;
3208     }
3209 }
3210
3211
3212 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
3213     MpegEncContext * const s = &h->s;
3214     AVCodecContext * const avctx= s->avctx;
3215     int buf_index=0;
3216     H264Context *hx; ///< thread context
3217     int context_count = 0;
3218     int next_avc= h->is_avc ? 0 : buf_size;
3219
3220     h->max_contexts = (HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_SLICE)) ? avctx->thread_count : 1;
3221 #if 0
3222     int i;
3223     for(i=0; i<50; i++){
3224         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
3225     }
3226 #endif
3227     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
3228         h->current_slice = 0;
3229         if (!s->first_field)
3230             s->current_picture_ptr= NULL;
3231         ff_h264_reset_sei(h);
3232     }
3233
3234     for(;;){
3235         int consumed;
3236         int dst_length;
3237         int bit_length;
3238         const uint8_t *ptr;
3239         int i, nalsize = 0;
3240         int err;
3241
3242         if(buf_index >= next_avc) {
3243             if(buf_index >= buf_size) break;
3244             nalsize = 0;
3245             for(i = 0; i < h->nal_length_size; i++)
3246                 nalsize = (nalsize << 8) | buf[buf_index++];
3247             if(nalsize <= 0 || nalsize > buf_size - buf_index){
3248                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
3249                 break;
3250             }
3251             next_avc= buf_index + nalsize;
3252         } else {
3253             // start code prefix search
3254             for(; buf_index + 3 < next_avc; buf_index++){
3255                 // This should always succeed in the first iteration.
3256                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
3257                     break;
3258             }
3259
3260             if(buf_index+3 >= buf_size) break;
3261
3262             buf_index+=3;
3263             if(buf_index >= next_avc) continue;
3264         }
3265
3266         hx = h->thread_context[context_count];
3267
3268         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
3269         if (ptr==NULL || dst_length < 0){
3270             return -1;
3271         }
3272         i= buf_index + consumed;
3273         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
3274            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
3275             s->workaround_bugs |= FF_BUG_TRUNCATED;
3276
3277         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
3278         while(ptr[dst_length - 1] == 0 && dst_length > 0)
3279             dst_length--;
3280         }
3281         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
3282
3283         if(s->avctx->debug&FF_DEBUG_STARTCODE){
3284             av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", hx->nal_unit_type, buf_index, buf_size, dst_length);
3285         }
3286
3287         if (h->is_avc && (nalsize != consumed) && nalsize){
3288             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
3289         }
3290
3291         buf_index += consumed;
3292
3293         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
3294            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
3295             continue;
3296
3297       again:
3298         err = 0;
3299         switch(hx->nal_unit_type){
3300         case NAL_IDR_SLICE:
3301             if (h->nal_unit_type != NAL_IDR_SLICE) {
3302                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
3303                 return -1;
3304             }
3305             idr(h); //FIXME ensure we don't loose some frames if there is reordering
3306         case NAL_SLICE:
3307             init_get_bits(&hx->s.gb, ptr, bit_length);
3308             hx->intra_gb_ptr=
3309             hx->inter_gb_ptr= &hx->s.gb;
3310             hx->s.data_partitioning = 0;
3311
3312             if((err = decode_slice_header(hx, h)))
3313                break;
3314
3315             s->current_picture_ptr->key_frame |=
3316                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
3317                     (h->sei_recovery_frame_cnt >= 0);
3318
3319             if (h->current_slice == 1) {
3320                 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3321                     decode_postinit(h);
3322                 }
3323
3324                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
3325                     return -1;
3326                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3327                     ff_vdpau_h264_picture_start(s);
3328             }
3329
3330             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
3331                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3332                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
3333                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
3334                && avctx->skip_frame < AVDISCARD_ALL){
3335                 if(avctx->hwaccel) {
3336                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
3337                         return -1;
3338                 }else
3339                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
3340                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
3341                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
3342                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
3343                 }else
3344                     context_count++;
3345             }
3346             break;
3347         case NAL_DPA:
3348             init_get_bits(&hx->s.gb, ptr, bit_length);
3349             hx->intra_gb_ptr=
3350             hx->inter_gb_ptr= NULL;
3351
3352             if ((err = decode_slice_header(hx, h)) < 0)
3353                 break;
3354
3355             hx->s.data_partitioning = 1;
3356
3357             break;
3358         case NAL_DPB:
3359             init_get_bits(&hx->intra_gb, ptr, bit_length);
3360             hx->intra_gb_ptr= &hx->intra_gb;
3361             break;
3362         case NAL_DPC:
3363             init_get_bits(&hx->inter_gb, ptr, bit_length);
3364             hx->inter_gb_ptr= &hx->inter_gb;
3365
3366             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
3367                && s->context_initialized
3368                && s->hurry_up < 5
3369                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3370                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
3371                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
3372                && avctx->skip_frame < AVDISCARD_ALL)
3373                 context_count++;
3374             break;
3375         case NAL_SEI:
3376             init_get_bits(&s->gb, ptr, bit_length);
3377             ff_h264_decode_sei(h);
3378             break;
3379         case NAL_SPS:
3380             init_get_bits(&s->gb, ptr, bit_length);
3381             ff_h264_decode_seq_parameter_set(h);
3382
3383             if(s->flags& CODEC_FLAG_LOW_DELAY)
3384                 s->low_delay=1;
3385
3386             if(avctx->has_b_frames < 2)
3387                 avctx->has_b_frames= !s->low_delay;
3388             break;
3389         case NAL_PPS:
3390             init_get_bits(&s->gb, ptr, bit_length);
3391
3392             ff_h264_decode_picture_parameter_set(h, bit_length);
3393
3394             break;
3395         case NAL_AUD:
3396         case NAL_END_SEQUENCE:
3397         case NAL_END_STREAM:
3398         case NAL_FILLER_DATA:
3399         case NAL_SPS_EXT:
3400         case NAL_AUXILIARY_SLICE:
3401             break;
3402         default:
3403             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
3404         }
3405
3406         if(context_count == h->max_contexts) {
3407             execute_decode_slices(h, context_count);
3408             context_count = 0;
3409         }
3410
3411         if (err < 0)
3412             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
3413         else if(err == 1) {
3414             /* Slice could not be decoded in parallel mode, copy down
3415              * NAL unit stuff to context 0 and restart. Note that
3416              * rbsp_buffer is not transferred, but since we no longer
3417              * run in parallel mode this should not be an issue. */
3418             h->nal_unit_type = hx->nal_unit_type;
3419             h->nal_ref_idc   = hx->nal_ref_idc;
3420             hx = h;
3421             goto again;
3422         }
3423     }
3424     if(context_count)
3425         execute_decode_slices(h, context_count);
3426     return buf_index;
3427 }
3428
3429 /**
3430  * returns the number of bytes consumed for building the current frame
3431  */
3432 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
3433         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
3434         if(pos+10>buf_size) pos=buf_size; // oops ;)
3435
3436         return pos;
3437 }
3438
3439 static int decode_frame(AVCodecContext *avctx,
3440                              void *data, int *data_size,
3441                              AVPacket *avpkt)
3442 {
3443     const uint8_t *buf = avpkt->data;
3444     int buf_size = avpkt->size;
3445     H264Context *h = avctx->priv_data;
3446     MpegEncContext *s = &h->s;
3447     AVFrame *pict = data;
3448     int buf_index;
3449
3450     s->flags= avctx->flags;
3451     s->flags2= avctx->flags2;
3452
3453    /* end of stream, output what is still in the buffers */
3454  out:
3455     if (buf_size == 0) {
3456         Picture *out;
3457         int i, out_idx;
3458
3459         s->current_picture_ptr = NULL;
3460
3461 //FIXME factorize this with the output code below
3462         out = h->delayed_pic[0];
3463         out_idx = 0;
3464         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3465             if(h->delayed_pic[i]->poc < out->poc){
3466                 out = h->delayed_pic[i];
3467                 out_idx = i;
3468             }
3469
3470         for(i=out_idx; h->delayed_pic[i]; i++)
3471             h->delayed_pic[i] = h->delayed_pic[i+1];
3472
3473         if(out){
3474             *data_size = sizeof(AVFrame);
3475             *pict= *(AVFrame*)out;
3476         }
3477
3478         return 0;
3479     }
3480
3481     buf_index=decode_nal_units(h, buf, buf_size);
3482     if(buf_index < 0)
3483         return -1;
3484
3485     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
3486         buf_size = 0;
3487         goto out;
3488     }
3489
3490     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3491         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
3492         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3493         return -1;
3494     }
3495
3496     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3497
3498         if(s->flags2 & CODEC_FLAG2_CHUNKS) decode_postinit(h);
3499
3500         field_end(h, 0);
3501
3502         if (!h->next_output_pic) {
3503             /* Wait for second field. */
3504             *data_size = 0;
3505
3506         } else {
3507             *data_size = sizeof(AVFrame);
3508         &