Add the notion of pixel size in h264 related functions.
[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 Libav.
6  *
7  * Libav is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * Libav is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21
22 /**
23  * @file
24  * H.264 / AVC / MPEG4 part10 codec.
25  * @author Michael Niedermayer <michaelni@gmx.at>
26  */
27
28 #include "libavutil/imgutils.h"
29 #include "internal.h"
30 #include "dsputil.h"
31 #include "avcodec.h"
32 #include "mpegvideo.h"
33 #include "h264.h"
34 #include "h264data.h"
35 #include "h264_mvpred.h"
36 #include "golomb.h"
37 #include "mathops.h"
38 #include "rectangle.h"
39 #include "vdpau_internal.h"
40 #include "libavutil/avassert.h"
41
42 #include "cabac.h"
43
44 //#undef NDEBUG
45 #include <assert.h>
46
47 static const uint8_t rem6[52]={
48 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,
49 };
50
51 static const uint8_t div6[52]={
52 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,
53 };
54
55 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
56     PIX_FMT_DXVA2_VLD,
57     PIX_FMT_VAAPI_VLD,
58     PIX_FMT_YUVJ420P,
59     PIX_FMT_NONE
60 };
61
62 void ff_h264_write_back_intra_pred_mode(H264Context *h){
63     int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
64
65     AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
66     mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
67     mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
68     mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
69 }
70
71 /**
72  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
73  */
74 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
75     MpegEncContext * const s = &h->s;
76     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
77     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
78     int i;
79
80     if(!(h->top_samples_available&0x8000)){
81         for(i=0; i<4; i++){
82             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
83             if(status<0){
84                 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);
85                 return -1;
86             } else if(status){
87                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
88             }
89         }
90     }
91
92     if((h->left_samples_available&0x8888)!=0x8888){
93         static const int mask[4]={0x8000,0x2000,0x80,0x20};
94         for(i=0; i<4; i++){
95             if(!(h->left_samples_available&mask[i])){
96                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
97                 if(status<0){
98                     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);
99                     return -1;
100                 } else if(status){
101                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
102                 }
103             }
104         }
105     }
106
107     return 0;
108 } //FIXME cleanup like ff_h264_check_intra_pred_mode
109
110 /**
111  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
112  */
113 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
114     MpegEncContext * const s = &h->s;
115     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
116     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
117
118     if(mode > 6U) {
119         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);
120         return -1;
121     }
122
123     if(!(h->top_samples_available&0x8000)){
124         mode= top[ mode ];
125         if(mode<0){
126             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);
127             return -1;
128         }
129     }
130
131     if((h->left_samples_available&0x8080) != 0x8080){
132         mode= left[ mode ];
133         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
134             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
135         }
136         if(mode<0){
137             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);
138             return -1;
139         }
140     }
141
142     return mode;
143 }
144
145 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
146     int i, si, di;
147     uint8_t *dst;
148     int bufidx;
149
150 //    src[0]&0x80;                //forbidden bit
151     h->nal_ref_idc= src[0]>>5;
152     h->nal_unit_type= src[0]&0x1F;
153
154     src++; length--;
155
156 #if HAVE_FAST_UNALIGNED
157 # if HAVE_FAST_64BIT
158 #   define RS 7
159     for(i=0; i+1<length; i+=9){
160         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
161 # else
162 #   define RS 3
163     for(i=0; i+1<length; i+=5){
164         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
165 # endif
166             continue;
167         if(i>0 && !src[i]) i--;
168         while(src[i]) i++;
169 #else
170 #   define RS 0
171     for(i=0; i+1<length; i+=2){
172         if(src[i]) continue;
173         if(i>0 && src[i-1]==0) i--;
174 #endif
175         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
176             if(src[i+2]!=3){
177                 /* startcode, so we must be past the end */
178                 length=i;
179             }
180             break;
181         }
182         i-= RS;
183     }
184
185     if(i>=length-1){ //no escaped 0
186         *dst_length= length;
187         *consumed= length+1; //+1 for the header
188         return src;
189     }
190
191     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
192     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
193     dst= h->rbsp_buffer[bufidx];
194
195     if (dst == NULL){
196         return NULL;
197     }
198
199 //printf("decoding esc\n");
200     memcpy(dst, src, i);
201     si=di=i;
202     while(si+2<length){
203         //remove escapes (very rare 1:2^22)
204         if(src[si+2]>3){
205             dst[di++]= src[si++];
206             dst[di++]= src[si++];
207         }else if(src[si]==0 && src[si+1]==0){
208             if(src[si+2]==3){ //escape
209                 dst[di++]= 0;
210                 dst[di++]= 0;
211                 si+=3;
212                 continue;
213             }else //next start code
214                 goto nsc;
215         }
216
217         dst[di++]= src[si++];
218     }
219     while(si<length)
220         dst[di++]= src[si++];
221 nsc:
222
223     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
224
225     *dst_length= di;
226     *consumed= si + 1;//+1 for the header
227 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
228     return dst;
229 }
230
231 /**
232  * Identify the exact end of the bitstream
233  * @return the length of the trailing, or 0 if damaged
234  */
235 static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
236     int v= *src;
237     int r;
238
239     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
240
241     for(r=1; r<9; r++){
242         if(v&1) return r;
243         v>>=1;
244     }
245     return 0;
246 }
247
248 #if 0
249 /**
250  * DCT transforms the 16 dc values.
251  * @param qp quantization parameter ??? FIXME
252  */
253 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
254 //    const int qmul= dequant_coeff[qp][0];
255     int i;
256     int temp[16]; //FIXME check if this is a good idea
257     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
258     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
259
260     for(i=0; i<4; i++){
261         const int offset= y_offset[i];
262         const int z0= block[offset+stride*0] + block[offset+stride*4];
263         const int z1= block[offset+stride*0] - block[offset+stride*4];
264         const int z2= block[offset+stride*1] - block[offset+stride*5];
265         const int z3= block[offset+stride*1] + block[offset+stride*5];
266
267         temp[4*i+0]= z0+z3;
268         temp[4*i+1]= z1+z2;
269         temp[4*i+2]= z1-z2;
270         temp[4*i+3]= z0-z3;
271     }
272
273     for(i=0; i<4; i++){
274         const int offset= x_offset[i];
275         const int z0= temp[4*0+i] + temp[4*2+i];
276         const int z1= temp[4*0+i] - temp[4*2+i];
277         const int z2= temp[4*1+i] - temp[4*3+i];
278         const int z3= temp[4*1+i] + temp[4*3+i];
279
280         block[stride*0 +offset]= (z0 + z3)>>1;
281         block[stride*2 +offset]= (z1 + z2)>>1;
282         block[stride*8 +offset]= (z1 - z2)>>1;
283         block[stride*10+offset]= (z0 - z3)>>1;
284     }
285 }
286 #endif
287
288 #undef xStride
289 #undef stride
290
291 #if 0
292 static void chroma_dc_dct_c(DCTELEM *block){
293     const int stride= 16*2;
294     const int xStride= 16;
295     int a,b,c,d,e;
296
297     a= block[stride*0 + xStride*0];
298     b= block[stride*0 + xStride*1];
299     c= block[stride*1 + xStride*0];
300     d= block[stride*1 + xStride*1];
301
302     e= a-b;
303     a= a+b;
304     b= c-d;
305     c= c+d;
306
307     block[stride*0 + xStride*0]= (a+c);
308     block[stride*0 + xStride*1]= (e+b);
309     block[stride*1 + xStride*0]= (a-c);
310     block[stride*1 + xStride*1]= (e-b);
311 }
312 #endif
313
314 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
315                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
316                            int src_x_offset, int src_y_offset,
317                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
318                            int pixel_shift){
319     MpegEncContext * const s = &h->s;
320     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
321     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
322     const int luma_xy= (mx&3) + ((my&3)<<2);
323     uint8_t * src_y = pic->data[0] + ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
324     uint8_t * src_cb, * src_cr;
325     int extra_width= h->emu_edge_width;
326     int extra_height= h->emu_edge_height;
327     int emu=0;
328     const int full_mx= mx>>2;
329     const int full_my= my>>2;
330     const int pic_width  = 16*s->mb_width;
331     const int pic_height = 16*s->mb_height >> MB_FIELD;
332
333     if(mx&7) extra_width -= 3;
334     if(my&7) extra_height -= 3;
335
336     if(   full_mx < 0-extra_width
337        || full_my < 0-extra_height
338        || full_mx + 16/*FIXME*/ > pic_width + extra_width
339        || full_my + 16/*FIXME*/ > pic_height + extra_height){
340         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
341             src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
342         emu=1;
343     }
344
345     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
346     if(!square){
347         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
348     }
349
350     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
351
352     if(MB_FIELD){
353         // chroma offset when predicting from a field of opposite parity
354         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
355         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
356     }
357     src_cb= pic->data[1] + ((mx>>3) << pixel_shift) + (my>>3)*h->mb_uvlinesize;
358     src_cr= pic->data[2] + ((mx>>3) << pixel_shift) + (my>>3)*h->mb_uvlinesize;
359
360     if(emu){
361         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);
362             src_cb= s->edge_emu_buffer;
363     }
364     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
365
366     if(emu){
367         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);
368             src_cr= s->edge_emu_buffer;
369     }
370     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
371 }
372
373 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
374                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
375                            int x_offset, int y_offset,
376                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
377                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
378                            int list0, int list1, int pixel_shift){
379     MpegEncContext * const s = &h->s;
380     qpel_mc_func *qpix_op=  qpix_put;
381     h264_chroma_mc_func chroma_op= chroma_put;
382
383     dest_y  += (2*x_offset << pixel_shift) + 2*y_offset*h->  mb_linesize;
384     dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
385     dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
386     x_offset += 8*s->mb_x;
387     y_offset += 8*(s->mb_y >> MB_FIELD);
388
389     if(list0){
390         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
391         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
392                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
393                            qpix_op, chroma_op, pixel_shift);
394
395         qpix_op=  qpix_avg;
396         chroma_op= chroma_avg;
397     }
398
399     if(list1){
400         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
401         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
402                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
403                            qpix_op, chroma_op, pixel_shift);
404     }
405 }
406
407 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
408                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
409                            int x_offset, int y_offset,
410                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
411                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
412                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
413                            int list0, int list1, int pixel_shift){
414     MpegEncContext * const s = &h->s;
415
416     dest_y  += (2*x_offset << pixel_shift) + 2*y_offset*h->  mb_linesize;
417     dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
418     dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
419     x_offset += 8*s->mb_x;
420     y_offset += 8*(s->mb_y >> MB_FIELD);
421
422     if(list0 && list1){
423         /* don't optimize for luma-only case, since B-frames usually
424          * use implicit weights => chroma too. */
425         uint8_t *tmp_cb = s->obmc_scratchpad;
426         uint8_t *tmp_cr = s->obmc_scratchpad + (8 << pixel_shift);
427         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
428         int refn0 = h->ref_cache[0][ scan8[n] ];
429         int refn1 = h->ref_cache[1][ scan8[n] ];
430
431         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
432                     dest_y, dest_cb, dest_cr,
433                     x_offset, y_offset, qpix_put, chroma_put, pixel_shift);
434         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
435                     tmp_y, tmp_cb, tmp_cr,
436                     x_offset, y_offset, qpix_put, chroma_put, pixel_shift);
437
438         if(h->use_weight == 2){
439             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
440             int weight1 = 64 - weight0;
441             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
442             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
443             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
444         }else{
445             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
446                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
447                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
448             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
449                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
450                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
451             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
452                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
453                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
454         }
455     }else{
456         int list = list1 ? 1 : 0;
457         int refn = h->ref_cache[list][ scan8[n] ];
458         Picture *ref= &h->ref_list[list][refn];
459         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
460                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
461                     qpix_put, chroma_put, pixel_shift);
462
463         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
464                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
465         if(h->use_weight_chroma){
466             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
467                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
468             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
469                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
470         }
471     }
472 }
473
474 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
475                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
476                            int x_offset, int y_offset,
477                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
478                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
479                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
480                            int list0, int list1, int pixel_shift){
481     if((h->use_weight==2 && list0 && list1
482         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
483        || h->use_weight==1)
484         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
485                          x_offset, y_offset, qpix_put, chroma_put,
486                          weight_op[0], weight_op[3], weight_avg[0],
487                          weight_avg[3], list0, list1, pixel_shift);
488     else
489         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
490                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
491                     chroma_avg, list0, list1, pixel_shift);
492 }
493
494 static inline void prefetch_motion(H264Context *h, int list, int pixel_shift){
495     /* fetch pixels for estimated mv 4 macroblocks ahead
496      * optimized for 64byte cache lines */
497     MpegEncContext * const s = &h->s;
498     const int refn = h->ref_cache[list][scan8[0]];
499     if(refn >= 0){
500         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
501         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
502         uint8_t **src= h->ref_list[list][refn].data;
503         int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
504         s->dsp.prefetch(src[0]+off, s->linesize, 4);
505         off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
506         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
507     }
508 }
509
510 static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
511                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
512                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
513                       h264_weight_func *weight_op, h264_biweight_func *weight_avg,
514                       int pixel_shift){
515     MpegEncContext * const s = &h->s;
516     const int mb_xy= h->mb_xy;
517     const int mb_type= s->current_picture.mb_type[mb_xy];
518
519     assert(IS_INTER(mb_type));
520
521     prefetch_motion(h, 0, pixel_shift);
522
523     if(IS_16X16(mb_type)){
524         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
525                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
526                 weight_op, weight_avg,
527                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
528                 pixel_shift);
529     }else if(IS_16X8(mb_type)){
530         mc_part(h, 0, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
531                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
532                 &weight_op[1], &weight_avg[1],
533                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
534                 pixel_shift);
535         mc_part(h, 8, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
536                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
537                 &weight_op[1], &weight_avg[1],
538                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
539                 pixel_shift);
540     }else if(IS_8X16(mb_type)){
541         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
542                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
543                 &weight_op[2], &weight_avg[2],
544                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
545                 pixel_shift);
546         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
547                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
548                 &weight_op[2], &weight_avg[2],
549                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
550                 pixel_shift);
551     }else{
552         int i;
553
554         assert(IS_8X8(mb_type));
555
556         for(i=0; i<4; i++){
557             const int sub_mb_type= h->sub_mb_type[i];
558             const int n= 4*i;
559             int x_offset= (i&1)<<2;
560             int y_offset= (i&2)<<1;
561
562             if(IS_SUB_8X8(sub_mb_type)){
563                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
564                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
565                     &weight_op[3], &weight_avg[3],
566                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
567                     pixel_shift);
568             }else if(IS_SUB_8X4(sub_mb_type)){
569                 mc_part(h, n  , 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
570                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
571                     &weight_op[4], &weight_avg[4],
572                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
573                     pixel_shift);
574                 mc_part(h, n+2, 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
575                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
576                     &weight_op[4], &weight_avg[4],
577                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
578                     pixel_shift);
579             }else if(IS_SUB_4X8(sub_mb_type)){
580                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
581                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
582                     &weight_op[5], &weight_avg[5],
583                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
584                     pixel_shift);
585                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
586                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
587                     &weight_op[5], &weight_avg[5],
588                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
589                     pixel_shift);
590             }else{
591                 int j;
592                 assert(IS_SUB_4X4(sub_mb_type));
593                 for(j=0; j<4; j++){
594                     int sub_x_offset= x_offset + 2*(j&1);
595                     int sub_y_offset= y_offset +   (j&2);
596                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
597                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
598                         &weight_op[6], &weight_avg[6],
599                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
600                         pixel_shift);
601                 }
602             }
603         }
604     }
605
606     prefetch_motion(h, 1, pixel_shift);
607 }
608
609 #define hl_motion_fn(sh, bits) \
610 static av_always_inline void hl_motion_ ## bits(H264Context *h, \
611                                        uint8_t *dest_y, \
612                                        uint8_t *dest_cb, uint8_t *dest_cr, \
613                                        qpel_mc_func (*qpix_put)[16], \
614                                        h264_chroma_mc_func (*chroma_put), \
615                                        qpel_mc_func (*qpix_avg)[16], \
616                                        h264_chroma_mc_func (*chroma_avg), \
617                                        h264_weight_func *weight_op, \
618                                        h264_biweight_func *weight_avg) \
619 { \
620     hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put, \
621               qpix_avg, chroma_avg, weight_op, weight_avg, sh); \
622 }
623 hl_motion_fn(0, 8);
624 hl_motion_fn(1, 16);
625
626 static void free_tables(H264Context *h, int free_rbsp){
627     int i;
628     H264Context *hx;
629     av_freep(&h->intra4x4_pred_mode);
630     av_freep(&h->chroma_pred_mode_table);
631     av_freep(&h->cbp_table);
632     av_freep(&h->mvd_table[0]);
633     av_freep(&h->mvd_table[1]);
634     av_freep(&h->direct_table);
635     av_freep(&h->non_zero_count);
636     av_freep(&h->slice_table_base);
637     h->slice_table= NULL;
638     av_freep(&h->list_counts);
639
640     av_freep(&h->mb2b_xy);
641     av_freep(&h->mb2br_xy);
642
643     for(i = 0; i < MAX_THREADS; i++) {
644         hx = h->thread_context[i];
645         if(!hx) continue;
646         av_freep(&hx->top_borders[1]);
647         av_freep(&hx->top_borders[0]);
648         av_freep(&hx->s.obmc_scratchpad);
649         if (free_rbsp){
650             av_freep(&hx->rbsp_buffer[1]);
651             av_freep(&hx->rbsp_buffer[0]);
652             hx->rbsp_buffer_size[0] = 0;
653             hx->rbsp_buffer_size[1] = 0;
654         }
655         if (i) av_freep(&h->thread_context[i]);
656     }
657 }
658
659 static void init_dequant8_coeff_table(H264Context *h){
660     int i,q,x;
661     h->dequant8_coeff[0] = h->dequant8_buffer[0];
662     h->dequant8_coeff[1] = h->dequant8_buffer[1];
663
664     for(i=0; i<2; i++ ){
665         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
666             h->dequant8_coeff[1] = h->dequant8_buffer[0];
667             break;
668         }
669
670         for(q=0; q<52; q++){
671             int shift = div6[q];
672             int idx = rem6[q];
673             for(x=0; x<64; x++)
674                 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
675                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
676                     h->pps.scaling_matrix8[i][x]) << shift;
677         }
678     }
679 }
680
681 static void init_dequant4_coeff_table(H264Context *h){
682     int i,j,q,x;
683     for(i=0; i<6; i++ ){
684         h->dequant4_coeff[i] = h->dequant4_buffer[i];
685         for(j=0; j<i; j++){
686             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
687                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
688                 break;
689             }
690         }
691         if(j<i)
692             continue;
693
694         for(q=0; q<52; q++){
695             int shift = div6[q] + 2;
696             int idx = rem6[q];
697             for(x=0; x<16; x++)
698                 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
699                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
700                     h->pps.scaling_matrix4[i][x]) << shift;
701         }
702     }
703 }
704
705 static void init_dequant_tables(H264Context *h){
706     int i,x;
707     init_dequant4_coeff_table(h);
708     if(h->pps.transform_8x8_mode)
709         init_dequant8_coeff_table(h);
710     if(h->sps.transform_bypass){
711         for(i=0; i<6; i++)
712             for(x=0; x<16; x++)
713                 h->dequant4_coeff[i][0][x] = 1<<6;
714         if(h->pps.transform_8x8_mode)
715             for(i=0; i<2; i++)
716                 for(x=0; x<64; x++)
717                     h->dequant8_coeff[i][0][x] = 1<<6;
718     }
719 }
720
721
722 int ff_h264_alloc_tables(H264Context *h){
723     MpegEncContext * const s = &h->s;
724     const int big_mb_num= s->mb_stride * (s->mb_height+1);
725     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
726     int x,y;
727
728     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
729
730     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
731     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
732     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
733
734     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
735     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
736     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
737     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
738     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
739
740     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
741     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
742
743     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
744     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
745     for(y=0; y<s->mb_height; y++){
746         for(x=0; x<s->mb_width; x++){
747             const int mb_xy= x + y*s->mb_stride;
748             const int b_xy = 4*x + 4*y*h->b_stride;
749
750             h->mb2b_xy [mb_xy]= b_xy;
751             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
752         }
753     }
754
755     s->obmc_scratchpad = NULL;
756
757     if(!h->dequant4_coeff[0])
758         init_dequant_tables(h);
759
760     return 0;
761 fail:
762     free_tables(h, 1);
763     return -1;
764 }
765
766 /**
767  * Mimic alloc_tables(), but for every context thread.
768  */
769 static void clone_tables(H264Context *dst, H264Context *src, int i){
770     MpegEncContext * const s = &src->s;
771     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
772     dst->non_zero_count           = src->non_zero_count;
773     dst->slice_table              = src->slice_table;
774     dst->cbp_table                = src->cbp_table;
775     dst->mb2b_xy                  = src->mb2b_xy;
776     dst->mb2br_xy                 = src->mb2br_xy;
777     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
778     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
779     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
780     dst->direct_table             = src->direct_table;
781     dst->list_counts              = src->list_counts;
782
783     dst->s.obmc_scratchpad = NULL;
784     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
785 }
786
787 /**
788  * Init context
789  * Allocate buffers which are not shared amongst multiple threads.
790  */
791 static int context_init(H264Context *h){
792     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t)*2, fail)
793     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t)*2, fail)
794
795     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
796     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
797
798     return 0;
799 fail:
800     return -1; // free_tables will clean up for us
801 }
802
803 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
804
805 static av_cold void common_init(H264Context *h){
806     MpegEncContext * const s = &h->s;
807
808     s->width = s->avctx->width;
809     s->height = s->avctx->height;
810     s->codec_id= s->avctx->codec->id;
811
812     ff_h264dsp_init(&h->h264dsp);
813     ff_h264_pred_init(&h->hpc, s->codec_id);
814
815     h->dequant_coeff_pps= -1;
816     s->unrestricted_mv=1;
817     s->decode=1; //FIXME
818
819     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
820
821     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
822     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
823 }
824
825 int ff_h264_decode_extradata(H264Context *h)
826 {
827     AVCodecContext *avctx = h->s.avctx;
828
829     if(*(char *)avctx->extradata == 1){
830         int i, cnt, nalsize;
831         unsigned char *p = avctx->extradata;
832
833         h->is_avc = 1;
834
835         if(avctx->extradata_size < 7) {
836             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
837             return -1;
838         }
839         /* sps and pps in the avcC always have length coded with 2 bytes,
840            so put a fake nal_length_size = 2 while parsing them */
841         h->nal_length_size = 2;
842         // Decode sps from avcC
843         cnt = *(p+5) & 0x1f; // Number of sps
844         p += 6;
845         for (i = 0; i < cnt; i++) {
846             nalsize = AV_RB16(p) + 2;
847             if(decode_nal_units(h, p, nalsize) < 0) {
848                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
849                 return -1;
850             }
851             p += nalsize;
852         }
853         // Decode pps from avcC
854         cnt = *(p++); // Number of pps
855         for (i = 0; i < cnt; i++) {
856             nalsize = AV_RB16(p) + 2;
857             if(decode_nal_units(h, p, nalsize)  != nalsize) {
858                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
859                 return -1;
860             }
861             p += nalsize;
862         }
863         // Now store right nal length size, that will be use to parse all other nals
864         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
865     } else {
866         h->is_avc = 0;
867         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
868             return -1;
869     }
870     return 0;
871 }
872
873 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
874     H264Context *h= avctx->priv_data;
875     MpegEncContext * const s = &h->s;
876
877     MPV_decode_defaults(s);
878
879     s->avctx = avctx;
880     common_init(h);
881
882     s->out_format = FMT_H264;
883     s->workaround_bugs= avctx->workaround_bugs;
884
885     // set defaults
886 //    s->decode_mb= ff_h263_decode_mb;
887     s->quarter_sample = 1;
888     if(!avctx->has_b_frames)
889     s->low_delay= 1;
890
891     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
892
893     ff_h264_decode_init_vlc();
894
895     h->pixel_shift = 0;
896
897     h->thread_context[0] = h;
898     h->outputed_poc = INT_MIN;
899     h->prev_poc_msb= 1<<16;
900     h->x264_build = -1;
901     ff_h264_reset_sei(h);
902     if(avctx->codec_id == CODEC_ID_H264){
903         if(avctx->ticks_per_frame == 1){
904             s->avctx->time_base.den *=2;
905         }
906         avctx->ticks_per_frame = 2;
907     }
908
909     if(avctx->extradata_size > 0 && avctx->extradata &&
910         ff_h264_decode_extradata(h))
911         return -1;
912
913     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
914         s->avctx->has_b_frames = h->sps.num_reorder_frames;
915         s->low_delay = 0;
916     }
917
918     return 0;
919 }
920
921 int ff_h264_frame_start(H264Context *h){
922     MpegEncContext * const s = &h->s;
923     int i;
924     const int pixel_shift = h->pixel_shift;
925
926     if(MPV_frame_start(s, s->avctx) < 0)
927         return -1;
928     ff_er_frame_start(s);
929     /*
930      * MPV_frame_start uses pict_type to derive key_frame.
931      * This is incorrect for H.264; IDR markings must be used.
932      * Zero here; IDR markings per slice in frame or fields are ORed in later.
933      * See decode_nal_units().
934      */
935     s->current_picture_ptr->key_frame= 0;
936     s->current_picture_ptr->mmco_reset= 0;
937
938     assert(s->linesize && s->uvlinesize);
939
940     for(i=0; i<16; i++){
941         h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
942         h->block_offset[24+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
943     }
944     for(i=0; i<4; i++){
945         h->block_offset[16+i]=
946         h->block_offset[20+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
947         h->block_offset[24+16+i]=
948         h->block_offset[24+20+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
949     }
950
951     /* can't be in alloc_tables because linesize isn't known there.
952      * FIXME: redo bipred weight to not require extra buffer? */
953     for(i = 0; i < s->avctx->thread_count; i++)
954         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
955             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
956
957     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
958     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
959
960 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
961
962     // We mark the current picture as non-reference after allocating it, so
963     // that if we break out due to an error it can be released automatically
964     // in the next MPV_frame_start().
965     // SVQ3 as well as most other codecs have only last/next/current and thus
966     // get released even with set reference, besides SVQ3 and others do not
967     // mark frames as reference later "naturally".
968     if(s->codec_id != CODEC_ID_SVQ3)
969         s->current_picture_ptr->reference= 0;
970
971     s->current_picture_ptr->field_poc[0]=
972     s->current_picture_ptr->field_poc[1]= INT_MAX;
973     assert(s->current_picture_ptr->long_ref==0);
974
975     return 0;
976 }
977
978 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){
979     MpegEncContext * const s = &h->s;
980     uint8_t *top_border;
981     int top_idx = 1;
982     const int pixel_shift = h->pixel_shift;
983
984     src_y  -=   linesize;
985     src_cb -= uvlinesize;
986     src_cr -= uvlinesize;
987
988     if(!simple && FRAME_MBAFF){
989         if(s->mb_y&1){
990             if(!MB_MBAFF){
991                 top_border = h->top_borders[0][s->mb_x];
992                 AV_COPY128(top_border, src_y + 15*linesize);
993                 if (pixel_shift)
994                     AV_COPY128(top_border+16, src_y+15*linesize+16);
995                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
996                     if (pixel_shift) {
997                         AV_COPY128(top_border+32, src_cb+7*uvlinesize);
998                         AV_COPY128(top_border+48, src_cr+7*uvlinesize);
999                     } else {
1000                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1001                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1002                     }
1003                 }
1004             }
1005         }else if(MB_MBAFF){
1006             top_idx = 0;
1007         }else
1008             return;
1009     }
1010
1011     top_border = h->top_borders[top_idx][s->mb_x];
1012     // There are two lines saved, the line above the the top macroblock of a pair,
1013     // and the line above the bottom macroblock
1014     AV_COPY128(top_border, src_y + 16*linesize);
1015     if (pixel_shift)
1016         AV_COPY128(top_border+16, src_y+16*linesize+16);
1017
1018     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1019         if (pixel_shift) {
1020             AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1021             AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1022         } else {
1023         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1024         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1025         }
1026     }
1027 }
1028
1029 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1030                                   uint8_t *src_cb, uint8_t *src_cr,
1031                                   int linesize, int uvlinesize,
1032                                   int xchg, int simple, int pixel_shift){
1033     MpegEncContext * const s = &h->s;
1034     int deblock_left;
1035     int deblock_top;
1036     int top_idx = 1;
1037     uint8_t *top_border_m1;
1038     uint8_t *top_border;
1039
1040     if(!simple && FRAME_MBAFF){
1041         if(s->mb_y&1){
1042             if(!MB_MBAFF)
1043                 return;
1044         }else{
1045             top_idx = MB_MBAFF ? 0 : 1;
1046         }
1047     }
1048
1049     if(h->deblocking_filter == 2) {
1050         deblock_left = h->left_type[0];
1051         deblock_top  = h->top_type;
1052     } else {
1053         deblock_left = (s->mb_x > 0);
1054         deblock_top =  (s->mb_y > !!MB_FIELD);
1055     }
1056
1057     src_y  -=   linesize + 1 + pixel_shift;
1058     src_cb -= uvlinesize + 1 + pixel_shift;
1059     src_cr -= uvlinesize + 1 + pixel_shift;
1060
1061     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1062     top_border    = h->top_borders[top_idx][s->mb_x];
1063
1064 #define XCHG(a,b,xchg)\
1065     if (pixel_shift) {\
1066         if (xchg) {\
1067             AV_SWAP64(b+0,a+0);\
1068             AV_SWAP64(b+8,a+8);\
1069         } else {\
1070             AV_COPY128(b,a); \
1071         }\
1072     } else \
1073 if (xchg) AV_SWAP64(b,a);\
1074 else      AV_COPY64(b,a);
1075
1076     if(deblock_top){
1077         if(deblock_left){
1078             XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
1079         }
1080         XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1081         XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1082         if(s->mb_x+1 < s->mb_width){
1083             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
1084         }
1085     }
1086     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1087         if(deblock_top){
1088             if(deblock_left){
1089                 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1090                 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1091             }
1092             XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1093             XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
1094         }
1095     }
1096 }
1097
1098 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1099     if (high_bit_depth) {
1100         return AV_RN32A(((int32_t*)mb) + index);
1101     } else
1102         return AV_RN16A(mb + index);
1103 }
1104
1105 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1106     if (high_bit_depth) {
1107         AV_WN32A(((int32_t*)mb) + index, value);
1108     } else
1109         AV_WN16A(mb + index, value);
1110 }
1111
1112 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift){
1113     MpegEncContext * const s = &h->s;
1114     const int mb_x= s->mb_x;
1115     const int mb_y= s->mb_y;
1116     const int mb_xy= h->mb_xy;
1117     const int mb_type= s->current_picture.mb_type[mb_xy];
1118     uint8_t  *dest_y, *dest_cb, *dest_cr;
1119     int linesize, uvlinesize /*dct_offset*/;
1120     int i;
1121     int *block_offset = &h->block_offset[0];
1122     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1123     /* is_h264 should always be true if SVQ3 is disabled. */
1124     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1125     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1126     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1127
1128     dest_y  = s->current_picture.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
1129     dest_cb = s->current_picture.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
1130     dest_cr = s->current_picture.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
1131
1132     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1133     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
1134
1135     h->list_counts[mb_xy]= h->list_count;
1136
1137     if (!simple && MB_FIELD) {
1138         linesize   = h->mb_linesize   = s->linesize * 2;
1139         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1140         block_offset = &h->block_offset[24];
1141         if(mb_y&1){ //FIXME move out of this function?
1142             dest_y -= s->linesize*15;
1143             dest_cb-= s->uvlinesize*7;
1144             dest_cr-= s->uvlinesize*7;
1145         }
1146         if(FRAME_MBAFF) {
1147             int list;
1148             for(list=0; list<h->list_count; list++){
1149                 if(!USES_LIST(mb_type, list))
1150                     continue;
1151                 if(IS_16X16(mb_type)){
1152                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1153                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1154                 }else{
1155                     for(i=0; i<16; i+=4){
1156                         int ref = h->ref_cache[list][scan8[i]];
1157                         if(ref >= 0)
1158                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1159                     }
1160                 }
1161             }
1162         }
1163     } else {
1164         linesize   = h->mb_linesize   = s->linesize;
1165         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1166 //        dct_offset = s->linesize * 16;
1167     }
1168
1169     if (!simple && IS_INTRA_PCM(mb_type)) {
1170         if (pixel_shift) {
1171             const int bit_depth = h->sps.bit_depth_luma;
1172             int j;
1173             GetBitContext gb;
1174             init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1175
1176             for (i = 0; i < 16; i++) {
1177                 uint16_t *tmp_y  = (uint16_t*)(dest_y  + i*linesize);
1178                 for (j = 0; j < 16; j++)
1179                     tmp_y[j] = get_bits(&gb, bit_depth);
1180             }
1181             for (i = 0; i < 8; i++) {
1182                 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1183                 for (j = 0; j < 8; j++)
1184                     tmp_cb[j] = get_bits(&gb, bit_depth);
1185             }
1186             for (i = 0; i < 8; i++) {
1187                 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1188                 for (j = 0; j < 8; j++)
1189                     tmp_cr[j] = get_bits(&gb, bit_depth);
1190             }
1191         } else {
1192         for (i=0; i<16; i++) {
1193             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1194         }
1195         for (i=0; i<8; i++) {
1196             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1197             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1198         }
1199         }
1200     } else {
1201         if(IS_INTRA(mb_type)){
1202             if(h->deblocking_filter)
1203                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple, pixel_shift);
1204
1205             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1206                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1207                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1208             }
1209
1210             if(IS_INTRA4x4(mb_type)){
1211                 if(simple || !s->encoding){
1212                     if(IS_8x8DCT(mb_type)){
1213                         if(transform_bypass){
1214                             idct_dc_add =
1215                             idct_add    = s->dsp.add_pixels8;
1216                         }else{
1217                             idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1218                             idct_add    = h->h264dsp.h264_idct8_add;
1219                         }
1220                         for(i=0; i<16; i+=4){
1221                             uint8_t * const ptr= dest_y + block_offset[i];
1222                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1223                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1224                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16 << pixel_shift), linesize);
1225                             }else{
1226                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1227                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1228                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1229                                 if(nnz){
1230                                     if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16))
1231                                         idct_dc_add(ptr, h->mb + (i*16 << pixel_shift), linesize);
1232                                     else
1233                                         idct_add   (ptr, h->mb + (i*16 << pixel_shift), linesize);
1234                                 }
1235                             }
1236                         }
1237                     }else{
1238                         if(transform_bypass){
1239                             idct_dc_add =
1240                             idct_add    = s->dsp.add_pixels4;
1241                         }else{
1242                             idct_dc_add = h->h264dsp.h264_idct_dc_add;
1243                             idct_add    = h->h264dsp.h264_idct_add;
1244                         }
1245                         for(i=0; i<16; i++){
1246                             uint8_t * const ptr= dest_y + block_offset[i];
1247                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1248
1249                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1250                                 h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16 << pixel_shift), linesize);
1251                             }else{
1252                                 uint8_t *topright;
1253                                 int nnz, tr;
1254                                 uint64_t tr_high;
1255                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1256                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1257                                     assert(mb_y || linesize <= block_offset[i]);
1258                                     if(!topright_avail){
1259                                         if (pixel_shift) {
1260                                             tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1261                                             topright= (uint8_t*) &tr_high;
1262                                         } else {
1263                                         tr= ptr[3 - linesize]*0x01010101;
1264                                         topright= (uint8_t*) &tr;
1265                                         }
1266                                     }else
1267                                         topright= ptr + (4 << pixel_shift) - linesize;
1268                                 }else
1269                                     topright= NULL;
1270
1271                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1272                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1273                                 if(nnz){
1274                                     if(is_h264){
1275                                         if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16))
1276                                             idct_dc_add(ptr, h->mb + (i*16 << pixel_shift), linesize);
1277                                         else
1278                                             idct_add   (ptr, h->mb + (i*16 << pixel_shift), linesize);
1279                                     }else
1280                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1281                                 }
1282                             }
1283                         }
1284                     }
1285                 }
1286             }else{
1287                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1288                 if(is_h264){
1289                     if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX] ]){
1290                         if(!transform_bypass)
1291                             h->h264dsp.h264_luma_dc_dequant_idct(h->mb, h->mb_luma_dc, h->dequant4_coeff[0][s->qscale][0]);
1292                         else{
1293                             static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1294                                                                     8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1295                             for(i = 0; i < 16; i++)
1296                                 dctcoef_set(h->mb, pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc, pixel_shift, i));
1297                         }
1298                     }
1299                 }else
1300                     ff_svq3_luma_dc_dequant_idct_c(h->mb, h->mb_luma_dc, s->qscale);
1301             }
1302             if(h->deblocking_filter)
1303                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple, pixel_shift);
1304         }else if(is_h264){
1305             if (pixel_shift) {
1306                 hl_motion_16(h, dest_y, dest_cb, dest_cr,
1307                              s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1308                              s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1309                              h->h264dsp.weight_h264_pixels_tab,
1310                              h->h264dsp.biweight_h264_pixels_tab);
1311             } else
1312                 hl_motion_8(h, dest_y, dest_cb, dest_cr,
1313                             s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1314                             s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1315                             h->h264dsp.weight_h264_pixels_tab,
1316                             h->h264dsp.biweight_h264_pixels_tab);
1317         }
1318
1319
1320         if(!IS_INTRA4x4(mb_type)){
1321             if(is_h264){
1322                 if(IS_INTRA16x16(mb_type)){
1323                     if(transform_bypass){
1324                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1325                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1326                         }else{
1327                             for(i=0; i<16; i++){
1328                                 if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
1329                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16 << pixel_shift), linesize);
1330                             }
1331                         }
1332                     }else{
1333                          h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1334                     }
1335                 }else if(h->cbp&15){
1336                     if(transform_bypass){
1337                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1338                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1339                         for(i=0; i<16; i+=di){
1340                             if(h->non_zero_count_cache[ scan8[i] ]){
1341                                 idct_add(dest_y + block_offset[i], h->mb + (i*16 << pixel_shift), linesize);
1342                             }
1343                         }
1344                     }else{
1345                         if(IS_8x8DCT(mb_type)){
1346                             h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1347                         }else{
1348                             h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1349                         }
1350                     }
1351                 }
1352             }else{
1353                 for(i=0; i<16; i++){
1354                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1355                         uint8_t * const ptr= dest_y + block_offset[i];
1356                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1357                     }
1358                 }
1359             }
1360         }
1361
1362         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1363             uint8_t *dest[2] = {dest_cb, dest_cr};
1364             if(transform_bypass){
1365                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1366                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16 << pixel_shift), uvlinesize);
1367                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + (20*16 << pixel_shift), uvlinesize);
1368                 }else{
1369                     idct_add = s->dsp.add_pixels4;
1370                     for(i=16; i<16+8; i++){
1371                         if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
1372                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
1373                     }
1374                 }
1375             }else{
1376                 if(is_h264){
1377                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
1378                         h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16 << pixel_shift)       , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1379                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
1380                         h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + ((16*16+4*16) << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1381                     h->h264dsp.h264_idct_add8(dest, block_offset,
1382                                               h->mb, uvlinesize,
1383                                               h->non_zero_count_cache);
1384                 }else{
1385                     h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1386                     h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1387                     for(i=16; i<16+8; i++){
1388                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1389                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1390                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1391                         }
1392                     }
1393                 }
1394             }
1395         }
1396     }
1397     if(h->cbp || IS_INTRA(mb_type))
1398         s->dsp.clear_blocks(h->mb);
1399 }
1400
1401 /**
1402  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1403  */
1404 #define hl_decode_mb_simple(sh, bits) \
1405 static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
1406     hl_decode_mb_internal(h, 1, sh); \
1407 }
1408 hl_decode_mb_simple(0, 8);
1409 hl_decode_mb_simple(1, 16);
1410
1411 /**
1412  * Process a macroblock; this handles edge cases, such as interlacing.
1413  */
1414 static void av_noinline hl_decode_mb_complex(H264Context *h){
1415     hl_decode_mb_internal(h, 0, h->pixel_shift);
1416 }
1417
1418 void ff_h264_hl_decode_mb(H264Context *h){
1419     MpegEncContext * const s = &h->s;
1420     const int mb_xy= h->mb_xy;
1421     const int mb_type= s->current_picture.mb_type[mb_xy];
1422     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1423
1424     if (is_complex) {
1425         hl_decode_mb_complex(h);
1426     } else if (h->pixel_shift) {
1427         hl_decode_mb_simple_16(h);
1428     } else
1429         hl_decode_mb_simple_8(h);
1430 }
1431
1432 static int pred_weight_table(H264Context *h){
1433     MpegEncContext * const s = &h->s;
1434     int list, i;
1435     int luma_def, chroma_def;
1436
1437     h->use_weight= 0;
1438     h->use_weight_chroma= 0;
1439     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1440     if(CHROMA)
1441         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1442     luma_def = 1<<h->luma_log2_weight_denom;
1443     chroma_def = 1<<h->chroma_log2_weight_denom;
1444
1445     for(list=0; list<2; list++){
1446         h->luma_weight_flag[list]   = 0;
1447         h->chroma_weight_flag[list] = 0;
1448         for(i=0; i<h->ref_count[list]; i++){
1449             int luma_weight_flag, chroma_weight_flag;
1450
1451             luma_weight_flag= get_bits1(&s->gb);
1452             if(luma_weight_flag){
1453                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1454                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1455                 if(   h->luma_weight[i][list][0] != luma_def
1456                    || h->luma_weight[i][list][1] != 0) {
1457                     h->use_weight= 1;
1458                     h->luma_weight_flag[list]= 1;
1459                 }
1460             }else{
1461                 h->luma_weight[i][list][0]= luma_def;
1462                 h->luma_weight[i][list][1]= 0;
1463             }
1464
1465             if(CHROMA){
1466                 chroma_weight_flag= get_bits1(&s->gb);
1467                 if(chroma_weight_flag){
1468                     int j;
1469                     for(j=0; j<2; j++){
1470                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1471                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1472                         if(   h->chroma_weight[i][list][j][0] != chroma_def
1473                            || h->chroma_weight[i][list][j][1] != 0) {
1474                             h->use_weight_chroma= 1;
1475                             h->chroma_weight_flag[list]= 1;
1476                         }
1477                     }
1478                 }else{
1479                     int j;
1480                     for(j=0; j<2; j++){
1481                         h->chroma_weight[i][list][j][0]= chroma_def;
1482                         h->chroma_weight[i][list][j][1]= 0;
1483                     }
1484                 }
1485             }
1486         }
1487         if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
1488     }
1489     h->use_weight= h->use_weight || h->use_weight_chroma;
1490     return 0;
1491 }
1492
1493 /**
1494  * Initialize implicit_weight table.
1495  * @param field  0/1 initialize the weight for interlaced MBAFF
1496  *                -1 initializes the rest
1497  */
1498 static void implicit_weight_table(H264Context *h, int field){
1499     MpegEncContext * const s = &h->s;
1500     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1501
1502     for (i = 0; i < 2; i++) {
1503         h->luma_weight_flag[i]   = 0;
1504         h->chroma_weight_flag[i] = 0;
1505     }
1506
1507     if(field < 0){
1508         cur_poc = s->current_picture_ptr->poc;
1509     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1510        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1511         h->use_weight= 0;
1512         h->use_weight_chroma= 0;
1513         return;
1514     }
1515         ref_start= 0;
1516         ref_count0= h->ref_count[0];
1517         ref_count1= h->ref_count[1];
1518     }else{
1519         cur_poc = s->current_picture_ptr->field_poc[field];
1520         ref_start= 16;
1521         ref_count0= 16+2*h->ref_count[0];
1522         ref_count1= 16+2*h->ref_count[1];
1523     }
1524
1525     h->use_weight= 2;
1526     h->use_weight_chroma= 2;
1527     h->luma_log2_weight_denom= 5;
1528     h->chroma_log2_weight_denom= 5;
1529
1530     for(ref0=ref_start; ref0 < ref_count0; ref0++){
1531         int poc0 = h->ref_list[0][ref0].poc;
1532         for(ref1=ref_start; ref1 < ref_count1; ref1++){
1533             int poc1 = h->ref_list[1][ref1].poc;
1534             int td = av_clip(poc1 - poc0, -128, 127);
1535             int w= 32;
1536             if(td){
1537                 int tb = av_clip(cur_poc - poc0, -128, 127);
1538                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1539                 int dist_scale_factor = (tb*tx + 32) >> 8;
1540                 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1541                     w = 64 - dist_scale_factor;
1542             }
1543             if(field<0){
1544                 h->implicit_weight[ref0][ref1][0]=
1545                 h->implicit_weight[ref0][ref1][1]= w;
1546             }else{
1547                 h->implicit_weight[ref0][ref1][field]=w;
1548             }
1549         }
1550     }
1551 }
1552
1553 /**
1554  * instantaneous decoder refresh.
1555  */
1556 static void idr(H264Context *h){
1557     ff_h264_remove_all_refs(h);
1558     h->prev_frame_num= 0;
1559     h->prev_frame_num_offset= 0;
1560     h->prev_poc_msb=
1561     h->prev_poc_lsb= 0;
1562 }
1563
1564 /* forget old pics after a seek */
1565 static void flush_dpb(AVCodecContext *avctx){
1566     H264Context *h= avctx->priv_data;
1567     int i;
1568     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1569         if(h->delayed_pic[i])
1570             h->delayed_pic[i]->reference= 0;
1571         h->delayed_pic[i]= NULL;
1572     }
1573     h->outputed_poc= INT_MIN;
1574     h->prev_interlaced_frame = 1;
1575     idr(h);
1576     if(h->s.current_picture_ptr)
1577         h->s.current_picture_ptr->reference= 0;
1578     h->s.first_field= 0;
1579     ff_h264_reset_sei(h);
1580     ff_mpeg_flush(avctx);
1581 }
1582
1583 static int init_poc(H264Context *h){
1584     MpegEncContext * const s = &h->s;
1585     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1586     int field_poc[2];
1587     Picture *cur = s->current_picture_ptr;
1588
1589     h->frame_num_offset= h->prev_frame_num_offset;
1590     if(h->frame_num < h->prev_frame_num)
1591         h->frame_num_offset += max_frame_num;
1592
1593     if(h->sps.poc_type==0){
1594         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1595
1596         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1597             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1598         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1599             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1600         else
1601             h->poc_msb = h->prev_poc_msb;
1602 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1603         field_poc[0] =
1604         field_poc[1] = h->poc_msb + h->poc_lsb;
1605         if(s->picture_structure == PICT_FRAME)
1606             field_poc[1] += h->delta_poc_bottom;
1607     }else if(h->sps.poc_type==1){
1608         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1609         int i;
1610
1611         if(h->sps.poc_cycle_length != 0)
1612             abs_frame_num = h->frame_num_offset + h->frame_num;
1613         else
1614             abs_frame_num = 0;
1615
1616         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1617             abs_frame_num--;
1618
1619         expected_delta_per_poc_cycle = 0;
1620         for(i=0; i < h->sps.poc_cycle_length; i++)
1621             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1622
1623         if(abs_frame_num > 0){
1624             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1625             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1626
1627             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1628             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1629                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1630         } else
1631             expectedpoc = 0;
1632
1633         if(h->nal_ref_idc == 0)
1634             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1635
1636         field_poc[0] = expectedpoc + h->delta_poc[0];
1637         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1638
1639         if(s->picture_structure == PICT_FRAME)
1640             field_poc[1] += h->delta_poc[1];
1641     }else{
1642         int poc= 2*(h->frame_num_offset + h->frame_num);
1643
1644         if(!h->nal_ref_idc)
1645             poc--;
1646
1647         field_poc[0]= poc;
1648         field_poc[1]= poc;
1649     }
1650
1651     if(s->picture_structure != PICT_BOTTOM_FIELD)
1652         s->current_picture_ptr->field_poc[0]= field_poc[0];
1653     if(s->picture_structure != PICT_TOP_FIELD)
1654         s->current_picture_ptr->field_poc[1]= field_poc[1];
1655     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1656
1657     return 0;
1658 }
1659
1660
1661 /**
1662  * initialize scan tables
1663  */
1664 static void init_scan_tables(H264Context *h){
1665     int i;
1666     for(i=0; i<16; i++){
1667 #define T(x) (x>>2) | ((x<<2) & 0xF)
1668         h->zigzag_scan[i] = T(zigzag_scan[i]);
1669         h-> field_scan[i] = T( field_scan[i]);
1670 #undef T
1671     }
1672     for(i=0; i<64; i++){
1673 #define T(x) (x>>3) | ((x&7)<<3)
1674         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1675         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1676         h->field_scan8x8[i]        = T(field_scan8x8[i]);
1677         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1678 #undef T
1679     }
1680     if(h->sps.transform_bypass){ //FIXME same ugly
1681         h->zigzag_scan_q0          = zigzag_scan;
1682         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1683         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1684         h->field_scan_q0           = field_scan;
1685         h->field_scan8x8_q0        = field_scan8x8;
1686         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1687     }else{
1688         h->zigzag_scan_q0          = h->zigzag_scan;
1689         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1690         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1691         h->field_scan_q0           = h->field_scan;
1692         h->field_scan8x8_q0        = h->field_scan8x8;
1693         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1694     }
1695 }
1696
1697 static void field_end(H264Context *h){
1698     MpegEncContext * const s = &h->s;
1699     AVCodecContext * const avctx= s->avctx;
1700     s->mb_y= 0;
1701
1702     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1703     s->current_picture_ptr->pict_type= s->pict_type;
1704
1705     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1706         ff_vdpau_h264_set_reference_frames(s);
1707
1708     if(!s->dropable) {
1709         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1710         h->prev_poc_msb= h->poc_msb;
1711         h->prev_poc_lsb= h->poc_lsb;
1712     }
1713     h->prev_frame_num_offset= h->frame_num_offset;
1714     h->prev_frame_num= h->frame_num;
1715
1716     if (avctx->hwaccel) {
1717         if (avctx->hwaccel->end_frame(avctx) < 0)
1718             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1719     }
1720
1721     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1722         ff_vdpau_h264_picture_complete(s);
1723
1724     /*
1725      * FIXME: Error handling code does not seem to support interlaced
1726      * when slices span multiple rows
1727      * The ff_er_add_slice calls don't work right for bottom
1728      * fields; they cause massive erroneous error concealing
1729      * Error marking covers both fields (top and bottom).
1730      * This causes a mismatched s->error_count
1731      * and a bad error table. Further, the error count goes to
1732      * INT_MAX when called for bottom field, because mb_y is
1733      * past end by one (callers fault) and resync_mb_y != 0
1734      * causes problems for the first MB line, too.
1735      */
1736     if (!FIELD_PICTURE)
1737         ff_er_frame_end(s);
1738
1739     MPV_frame_end(s);
1740
1741     h->current_slice=0;
1742 }
1743
1744 /**
1745  * Replicate H264 "master" context to thread contexts.
1746  */
1747 static void clone_slice(H264Context *dst, H264Context *src)
1748 {
1749     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1750     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1751     dst->s.current_picture      = src->s.current_picture;
1752     dst->s.linesize             = src->s.linesize;
1753     dst->s.uvlinesize           = src->s.uvlinesize;
1754     dst->s.first_field          = src->s.first_field;
1755
1756     dst->prev_poc_msb           = src->prev_poc_msb;
1757     dst->prev_poc_lsb           = src->prev_poc_lsb;
1758     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1759     dst->prev_frame_num         = src->prev_frame_num;
1760     dst->short_ref_count        = src->short_ref_count;
1761
1762     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1763     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1764     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1765     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1766
1767     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1768     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1769 }
1770
1771 /**
1772  * computes profile from profile_idc and constraint_set?_flags
1773  *
1774  * @param sps SPS
1775  *
1776  * @return profile as defined by FF_PROFILE_H264_*
1777  */
1778 int ff_h264_get_profile(SPS *sps)
1779 {
1780     int profile = sps->profile_idc;
1781
1782     switch(sps->profile_idc) {
1783     case FF_PROFILE_H264_BASELINE:
1784         // constraint_set1_flag set to 1
1785         profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
1786         break;
1787     case FF_PROFILE_H264_HIGH_10:
1788     case FF_PROFILE_H264_HIGH_422:
1789     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
1790         // constraint_set3_flag set to 1
1791         profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
1792         break;
1793     }
1794
1795     return profile;
1796 }
1797
1798 /**
1799  * decodes a slice header.
1800  * This will also call MPV_common_init() and frame_start() as needed.
1801  *
1802  * @param h h264context
1803  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1804  *
1805  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1806  */
1807 static int decode_slice_header(H264Context *h, H264Context *h0){
1808     MpegEncContext * const s = &h->s;
1809     MpegEncContext * const s0 = &h0->s;
1810     unsigned int first_mb_in_slice;
1811     unsigned int pps_id;
1812     int num_ref_idx_active_override_flag;
1813     unsigned int slice_type, tmp, i, j;
1814     int default_ref_list_done = 0;
1815     int last_pic_structure;
1816
1817     s->dropable= h->nal_ref_idc == 0;
1818
1819     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1820         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1821         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1822     }else{
1823         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1824         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1825     }
1826
1827     first_mb_in_slice= get_ue_golomb(&s->gb);
1828
1829     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1830         if(h0->current_slice && FIELD_PICTURE){
1831             field_end(h);
1832         }
1833
1834         h0->current_slice = 0;
1835         if (!s0->first_field)
1836             s->current_picture_ptr= NULL;
1837     }
1838
1839     slice_type= get_ue_golomb_31(&s->gb);
1840     if(slice_type > 9){
1841         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);
1842         return -1;
1843     }
1844     if(slice_type > 4){
1845         slice_type -= 5;
1846         h->slice_type_fixed=1;
1847     }else
1848         h->slice_type_fixed=0;
1849
1850     slice_type= golomb_to_pict_type[ slice_type ];
1851     if (slice_type == AV_PICTURE_TYPE_I
1852         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1853         default_ref_list_done = 1;
1854     }
1855     h->slice_type= slice_type;
1856     h->slice_type_nos= slice_type & 3;
1857
1858     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1859
1860     pps_id= get_ue_golomb(&s->gb);
1861     if(pps_id>=MAX_PPS_COUNT){
1862         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1863         return -1;
1864     }
1865     if(!h0->pps_buffers[pps_id]) {
1866         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1867         return -1;
1868     }
1869     h->pps= *h0->pps_buffers[pps_id];
1870
1871     if(!h0->sps_buffers[h->pps.sps_id]) {
1872         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1873         return -1;
1874     }
1875     h->sps = *h0->sps_buffers[h->pps.sps_id];
1876
1877     s->avctx->profile = ff_h264_get_profile(&h->sps);
1878     s->avctx->level   = h->sps.level_idc;
1879     s->avctx->refs    = h->sps.ref_frame_count;
1880
1881     if(h == h0 && h->dequant_coeff_pps != pps_id){
1882         h->dequant_coeff_pps = pps_id;
1883         init_dequant_tables(h);
1884     }
1885
1886     s->mb_width= h->sps.mb_width;
1887     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1888
1889     h->b_stride=  s->mb_width*4;
1890
1891     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1892     if(h->sps.frame_mbs_only_flag)
1893         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1894     else
1895         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 7);
1896
1897     if (s->context_initialized
1898         && (   s->width != s->avctx->width || s->height != s->avctx->height
1899             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
1900         if(h != h0)
1901             return -1;   // width / height changed during parallelized decoding
1902         free_tables(h, 0);
1903         flush_dpb(s->avctx);
1904         MPV_common_end(s);
1905     }
1906     if (!s->context_initialized) {
1907         if(h != h0)
1908             return -1;  // we cant (re-)initialize context during parallel decoding
1909
1910         avcodec_set_dimensions(s->avctx, s->width, s->height);
1911         s->avctx->sample_aspect_ratio= h->sps.sar;
1912         av_assert0(s->avctx->sample_aspect_ratio.den);
1913
1914         if(h->sps.video_signal_type_present_flag){
1915             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1916             if(h->sps.colour_description_present_flag){
1917                 s->avctx->color_primaries = h->sps.color_primaries;
1918                 s->avctx->color_trc       = h->sps.color_trc;
1919                 s->avctx->colorspace      = h->sps.colorspace;
1920             }
1921         }
1922
1923         if(h->sps.timing_info_present_flag){
1924             int64_t den= h->sps.time_scale;
1925             if(h->x264_build < 44U)
1926                 den *= 2;
1927             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1928                       h->sps.num_units_in_tick, den, 1<<30);
1929         }
1930         s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
1931                                                  s->avctx->codec->pix_fmts ?
1932                                                  s->avctx->codec->pix_fmts :
1933                                                  s->avctx->color_range == AVCOL_RANGE_JPEG ?
1934                                                  hwaccel_pixfmt_list_h264_jpeg_420 :
1935                                                  ff_hwaccel_pixfmt_list_420);
1936         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1937
1938         if (MPV_common_init(s) < 0)
1939             return -1;
1940         s->first_field = 0;
1941         h->prev_interlaced_frame = 1;
1942
1943         init_scan_tables(h);
1944         ff_h264_alloc_tables(h);
1945
1946         for(i = 1; i < s->avctx->thread_count; i++) {
1947             H264Context *c;
1948             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1949             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1950             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1951             c->h264dsp = h->h264dsp;
1952             c->sps = h->sps;
1953             c->pps = h->pps;
1954             init_scan_tables(c);
1955             clone_tables(c, h, i);
1956         }
1957
1958         for(i = 0; i < s->avctx->thread_count; i++)
1959             if(context_init(h->thread_context[i]) < 0)
1960                 return -1;
1961     }
1962
1963     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1964
1965     h->mb_mbaff = 0;
1966     h->mb_aff_frame = 0;
1967     last_pic_structure = s0->picture_structure;
1968     if(h->sps.frame_mbs_only_flag){
1969         s->picture_structure= PICT_FRAME;
1970     }else{
1971         if(get_bits1(&s->gb)) { //field_pic_flag
1972             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1973         } else {
1974             s->picture_structure= PICT_FRAME;
1975             h->mb_aff_frame = h->sps.mb_aff;
1976         }
1977     }
1978     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1979
1980     if(h0->current_slice == 0){
1981         while(h->frame_num !=  h->prev_frame_num &&
1982               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1983             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1984             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1985             if (ff_h264_frame_start(h) < 0)
1986                 return -1;
1987             h->prev_frame_num++;
1988             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1989             s->current_picture_ptr->frame_num= h->prev_frame_num;
1990             ff_generate_sliding_window_mmcos(h);
1991             ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1992             /* Error concealment: if a ref is missing, copy the previous ref in its place.
1993              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
1994              * about there being no actual duplicates.
1995              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
1996              * concealing a lost frame, this probably isn't noticable by comparison, but it should
1997              * be fixed. */
1998             if (h->short_ref_count) {
1999                 if (prev) {
2000                     av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
2001                                   (const uint8_t**)prev->data, prev->linesize,
2002                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2003                     h->short_ref[0]->poc = prev->poc+2;
2004                 }
2005                 h->short_ref[0]->frame_num = h->prev_frame_num;
2006             }
2007         }
2008
2009         /* See if we have a decoded first field looking for a pair... */
2010         if (s0->first_field) {
2011             assert(s0->current_picture_ptr);
2012             assert(s0->current_picture_ptr->data[0]);
2013             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
2014
2015             /* figure out if we have a complementary field pair */
2016             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2017                 /*
2018                  * Previous field is unmatched. Don't display it, but let it
2019                  * remain for reference if marked as such.
2020                  */
2021                 s0->current_picture_ptr = NULL;
2022                 s0->first_field = FIELD_PICTURE;
2023
2024             } else {
2025                 if (h->nal_ref_idc &&
2026                         s0->current_picture_ptr->reference &&
2027                         s0->current_picture_ptr->frame_num != h->frame_num) {
2028                     /*
2029                      * This and previous field were reference, but had
2030                      * different frame_nums. Consider this field first in
2031                      * pair. Throw away previous field except for reference
2032                      * purposes.
2033                      */
2034                     s0->first_field = 1;
2035                     s0->current_picture_ptr = NULL;
2036
2037                 } else {
2038                     /* Second field in complementary pair */
2039                     s0->first_field = 0;
2040                 }
2041             }
2042
2043         } else {
2044             /* Frame or first field in a potentially complementary pair */
2045             assert(!s0->current_picture_ptr);
2046             s0->first_field = FIELD_PICTURE;
2047         }
2048
2049         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
2050             s0->first_field = 0;
2051             return -1;
2052         }
2053     }
2054     if(h != h0)
2055         clone_slice(h, h0);
2056
2057     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2058
2059     assert(s->mb_num == s->mb_width * s->mb_height);
2060     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2061        first_mb_in_slice                    >= s->mb_num){
2062         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2063         return -1;
2064     }
2065     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2066     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2067     if (s->picture_structure == PICT_BOTTOM_FIELD)
2068         s->resync_mb_y = s->mb_y = s->mb_y + 1;
2069     assert(s->mb_y < s->mb_height);
2070
2071     if(s->picture_structure==PICT_FRAME){
2072         h->curr_pic_num=   h->frame_num;
2073         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2074     }else{
2075         h->curr_pic_num= 2*h->frame_num + 1;
2076         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2077     }
2078
2079     if(h->nal_unit_type == NAL_IDR_SLICE){
2080         get_ue_golomb(&s->gb); /* idr_pic_id */
2081     }
2082
2083     if(h->sps.poc_type==0){
2084         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2085
2086         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2087             h->delta_poc_bottom= get_se_golomb(&s->gb);
2088         }
2089     }
2090
2091     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2092         h->delta_poc[0]= get_se_golomb(&s->gb);
2093
2094         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2095             h->delta_poc[1]= get_se_golomb(&s->gb);
2096     }
2097
2098     init_poc(h);
2099
2100     if(h->pps.redundant_pic_cnt_present){
2101         h->redundant_pic_count= get_ue_golomb(&s->gb);
2102     }
2103
2104     //set defaults, might be overridden a few lines later
2105     h->ref_count[0]= h->pps.ref_count[0];
2106     h->ref_count[1]= h->pps.ref_count[1];
2107
2108     if(h->slice_type_nos != AV_PICTURE_TYPE_I){
2109         if(h->slice_type_nos == AV_PICTURE_TYPE_B){
2110             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2111         }
2112         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2113
2114         if(num_ref_idx_active_override_flag){
2115             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2116             if(h->slice_type_nos==AV_PICTURE_TYPE_B)
2117                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2118
2119             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2120                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2121                 h->ref_count[0]= h->ref_count[1]= 1;
2122                 return -1;
2123             }
2124         }
2125         if(h->slice_type_nos == AV_PICTURE_TYPE_B)
2126             h->list_count= 2;
2127         else
2128             h->list_count= 1;
2129     }else
2130         h->list_count= 0;
2131
2132     if(!default_ref_list_done){
2133         ff_h264_fill_default_ref_list(h);
2134     }
2135
2136     if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2137         return -1;
2138
2139     if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
2140         s->last_picture_ptr= &h->ref_list[0][0];
2141         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2142     }
2143     if(h->slice_type_nos==AV_PICTURE_TYPE_B){
2144         s->next_picture_ptr= &h->ref_list[1][0];
2145         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2146     }
2147
2148     if(   (h->pps.weighted_pred          && h->slice_type_nos == AV_PICTURE_TYPE_P )
2149        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
2150         pred_weight_table(h);
2151     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2152         implicit_weight_table(h, -1);
2153     }else {
2154         h->use_weight = 0;
2155         for (i = 0; i < 2; i++) {
2156             h->luma_weight_flag[i]   = 0;
2157             h->chroma_weight_flag[i] = 0;
2158         }
2159     }
2160
2161     if(h->nal_ref_idc)
2162         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2163
2164     if(FRAME_MBAFF){
2165         ff_h264_fill_mbaff_ref_list(h);
2166
2167         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2168             implicit_weight_table(h, 0);
2169             implicit_weight_table(h, 1);
2170         }
2171     }
2172
2173     if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
2174         ff_h264_direct_dist_scale_factor(h);
2175     ff_h264_direct_ref_list_init(h);
2176
2177     if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
2178         tmp = get_ue_golomb_31(&s->gb);
2179         if(tmp > 2){
2180             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2181             return -1;
2182         }
2183         h->cabac_init_idc= tmp;
2184     }
2185
2186     h->last_qscale_diff = 0;
2187     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2188     if(tmp>51){
2189         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2190         return -1;
2191     }
2192     s->qscale= tmp;
2193     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2194     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2195     //FIXME qscale / qp ... stuff
2196     if(h->slice_type == AV_PICTURE_TYPE_SP){
2197         get_bits1(&s->gb); /* sp_for_switch_flag */
2198     }
2199     if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
2200         get_se_golomb(&s->gb); /* slice_qs_delta */
2201     }
2202
2203     h->deblocking_filter = 1;
2204     h->slice_alpha_c0_offset = 52;
2205     h->slice_beta_offset = 52;
2206     if( h->pps.deblocking_filter_parameters_present ) {
2207         tmp= get_ue_golomb_31(&s->gb);
2208         if(tmp > 2){
2209             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2210             return -1;
2211         }
2212         h->deblocking_filter= tmp;
2213         if(h->deblocking_filter < 2)
2214             h->deblocking_filter^= 1; // 1<->0
2215
2216         if( h->deblocking_filter ) {
2217             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2218             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2219             if(   h->slice_alpha_c0_offset > 104U
2220                || h->slice_beta_offset     > 104U){
2221                 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);
2222                 return -1;
2223             }
2224         }
2225     }
2226
2227     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2228        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
2229        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == AV_PICTURE_TYPE_B)
2230        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2231         h->deblocking_filter= 0;
2232
2233     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2234         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2235             /* Cheat slightly for speed:
2236                Do not bother to deblock across slices. */
2237             h->deblocking_filter = 2;
2238         } else {
2239             h0->max_contexts = 1;
2240             if(!h0->single_decode_warning) {
2241                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2242                 h0->single_decode_warning = 1;
2243             }
2244             if(h != h0)
2245                 return 1; // deblocking switched inside frame
2246         }
2247     }
2248     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]);
2249
2250 #if 0 //FMO
2251     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2252         slice_group_change_cycle= get_bits(&s->gb, ?);
2253 #endif
2254
2255     h0->last_slice_type = slice_type;
2256     h->slice_num = ++h0->current_slice;
2257     if(h->slice_num >= MAX_SLICES){
2258         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2259     }
2260
2261     for(j=0; j<2; j++){
2262         int id_list[16];
2263         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2264         for(i=0; i<16; i++){
2265             id_list[i]= 60;
2266             if(h->ref_list[j][i].data[0]){
2267                 int k;
2268                 uint8_t *base= h->ref_list[j][i].base[0];
2269                 for(k=0; k<h->short_ref_count; k++)
2270                     if(h->short_ref[k]->base[0] == base){
2271                         id_list[i]= k;
2272                         break;
2273                     }
2274                 for(k=0; k<h->long_ref_count; k++)
2275                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2276                         id_list[i]= h->short_ref_count + k;
2277                         break;
2278                     }
2279             }
2280         }
2281
2282         ref2frm[0]=
2283         ref2frm[1]= -1;
2284         for(i=0; i<16; i++)
2285             ref2frm[i+2]= 4*id_list[i]
2286                           +(h->ref_list[j][i].reference&3);
2287         ref2frm[18+0]=
2288         ref2frm[18+1]= -1;
2289         for(i=16; i<48; i++)
2290             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2291                           +(h->ref_list[j][i].reference&3);
2292     }
2293
2294     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2295     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2296
2297     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2298         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",
2299                h->slice_num,
2300                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2301                first_mb_in_slice,
2302                av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2303                pps_id, h->frame_num,
2304                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2305                h->ref_count[0], h->ref_count[1],
2306                s->qscale,
2307                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2308                h->use_weight,
2309                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2310                h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2311                );
2312     }
2313
2314     return 0;
2315 }
2316
2317 int ff_h264_get_slice_type(const H264Context *h)
2318 {
2319     switch (h->slice_type) {
2320     case AV_PICTURE_TYPE_P:  return 0;
2321     case AV_PICTURE_TYPE_B:  return 1;
2322     case AV_PICTURE_TYPE_I:  return 2;
2323     case AV_PICTURE_TYPE_SP: return 3;
2324     case AV_PICTURE_TYPE_SI: return 4;
2325     default:         return -1;
2326     }
2327 }
2328
2329 /**
2330  *
2331  * @return non zero if the loop filter can be skiped
2332  */
2333 static int fill_filter_caches(H264Context *h, int mb_type){
2334     MpegEncContext * const s = &h->s;
2335     const int mb_xy= h->mb_xy;
2336     int top_xy, left_xy[2];
2337     int top_type, left_type[2];
2338
2339     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
2340
2341     //FIXME deblocking could skip the intra and nnz parts.
2342
2343     /* Wow, what a mess, why didn't they simplify the interlacing & intra
2344      * stuff, I can't imagine that these complex rules are worth it. */
2345
2346     left_xy[1] = left_xy[0] = mb_xy-1;
2347     if(FRAME_MBAFF){
2348         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2349         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
2350         if(s->mb_y&1){
2351             if (left_mb_field_flag != curr_mb_field_flag) {
2352                 left_xy[0] -= s->mb_stride;
2353             }
2354         }else{
2355             if(curr_mb_field_flag){
2356                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
2357             }
2358             if (left_mb_field_flag != curr_mb_field_flag) {
2359                 left_xy[1] += s->mb_stride;
2360             }
2361         }
2362     }
2363
2364     h->top_mb_xy = top_xy;
2365     h->left_mb_xy[0] = left_xy[0];
2366     h->left_mb_xy[1] = left_xy[1];
2367     {
2368         //for sufficiently low qp, filtering wouldn't do anything
2369         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2370         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2371         int qp = s->current_picture.qscale_table[mb_xy];
2372         if(qp <= qp_thresh
2373            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2374            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
2375             if(!FRAME_MBAFF)
2376                 return 1;
2377             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
2378                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
2379                 return 1;
2380         }
2381     }
2382
2383     top_type     = s->current_picture.mb_type[top_xy]    ;
2384     left_type[0] = s->current_picture.mb_type[left_xy[0]];
2385     left_type[1] = s->current_picture.mb_type[left_xy[1]];
2386     if(h->deblocking_filter == 2){
2387         if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
2388         if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2389     }else{
2390         if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
2391         if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2392     }
2393     h->top_type    = top_type    ;
2394     h->left_type[0]= left_type[0];
2395     h->left_type[1]= left_type[1];
2396
2397     if(IS_INTRA(mb_type))
2398         return 0;
2399
2400     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2401     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2402     AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2403     AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2404     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2405
2406     h->cbp= h->cbp_table[mb_xy];
2407
2408     {
2409         int list;
2410         for(list=0; list<h->list_count; list++){
2411             int8_t *ref;
2412             int y, b_stride;
2413             int16_t (*mv_dst)[2];
2414             int16_t (*mv_src)[2];
2415
2416             if(!USES_LIST(mb_type, list)){
2417                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2418                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2419                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2420                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2421                 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2422                 continue;
2423             }
2424
2425             ref = &s->current_picture.ref_index[list][4*mb_xy];
2426             {
2427                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2428                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2429                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2430                 ref += 2;
2431                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2432                 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2433             }
2434
2435             b_stride = h->b_stride;
2436             mv_dst   = &h->mv_cache[list][scan8[0]];
2437             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2438             for(y=0; y<4; y++){
2439                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2440             }
2441
2442         }
2443     }
2444
2445
2446 /*
2447 0 . T T. T T T T
2448 1 L . .L . . . .
2449 2 L . .L . . . .
2450 3 . T TL . . . .
2451 4 L . .L . . . .
2452 5 L . .. . . . .
2453 */
2454 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2455     if(top_type){
2456         AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2457     }
2458
2459     if(left_type[0]){
2460         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2461         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2462         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2463         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2464     }
2465
2466     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2467     if(!CABAC && h->pps.transform_8x8_mode){
2468         if(IS_8x8DCT(top_type)){
2469             h->non_zero_count_cache[4+8*0]=
2470             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2471             h->non_zero_count_cache[6+8*0]=
2472             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2473         }
2474         if(IS_8x8DCT(left_type[0])){
2475             h->non_zero_count_cache[3+8*1]=
2476             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2477         }
2478         if(IS_8x8DCT(left_type[1])){
2479             h->non_zero_count_cache[3+8*3]=
2480             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2481         }
2482
2483         if(IS_8x8DCT(mb_type)){
2484             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
2485             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
2486
2487             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2488             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2489
2490             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2491             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2492
2493             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2494             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2495         }
2496     }
2497
2498     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2499         int list;
2500         for(list=0; list<h->list_count; list++){
2501             if(USES_LIST(top_type, list)){
2502                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2503                 const int b8_xy= 4*top_xy + 2;
2504                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2505                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2506                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2507                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2508                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2509                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2510             }else{
2511                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2512                 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2513             }
2514
2515             if(!IS_INTERLACED(mb_type^left_type[0])){
2516                 if(USES_LIST(left_type[0], list)){
2517                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2518                     const int b8_xy= 4*left_xy[0] + 1;
2519                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2520                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2521                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2522                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2523                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2524                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
2525                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2526                     h->ref_cache[list][scan8[0] - 1 +16 ]=
2527                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2528                 }else{
2529                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2530                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2531                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2532                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2533                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
2534                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
2535                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
2536                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2537                 }
2538             }
2539         }
2540     }
2541
2542     return 0;
2543 }
2544
2545 static void loop_filter(H264Context *h){
2546     MpegEncContext * const s = &h->s;
2547     uint8_t  *dest_y, *dest_cb, *dest_cr;
2548     int linesize, uvlinesize, mb_x, mb_y;
2549     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2550     const int old_slice_type= h->slice_type;
2551     const int pixel_shift = h->pixel_shift;
2552
2553     if(h->deblocking_filter) {
2554         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2555             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2556                 int mb_xy, mb_type;
2557                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2558                 h->slice_num= h->slice_table[mb_xy];
2559                 mb_type= s->current_picture.mb_type[mb_xy];
2560                 h->list_count= h->list_counts[mb_xy];
2561
2562                 if(FRAME_MBAFF)
2563                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2564
2565                 s->mb_x= mb_x;
2566                 s->mb_y= mb_y;
2567                 dest_y  = s->current_picture.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
2568                 dest_cb = s->current_picture.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
2569                 dest_cr = s->current_picture.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
2570                     //FIXME simplify above
2571
2572                 if (MB_FIELD) {
2573                     linesize   = h->mb_linesize   = s->linesize * 2;
2574                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2575                     if(mb_y&1){ //FIXME move out of this function?
2576                         dest_y -= s->linesize*15;
2577                         dest_cb-= s->uvlinesize*7;
2578                         dest_cr-= s->uvlinesize*7;
2579                     }
2580                 } else {
2581                     linesize   = h->mb_linesize   = s->linesize;
2582                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2583                 }
2584                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2585                 if(fill_filter_caches(h, mb_type))
2586                     continue;
2587                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2588                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2589
2590                 if (FRAME_MBAFF) {
2591                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2592                 } else {
2593                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2594                 }
2595             }
2596         }
2597     }
2598     h->slice_type= old_slice_type;
2599     s->mb_x= 0;
2600     s->mb_y= end_mb_y - FRAME_MBAFF;
2601     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2602     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2603 }
2604
2605 static void predict_field_decoding_flag(H264Context *h){
2606     MpegEncContext * const s = &h->s;
2607     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2608     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2609                 ? s->current_picture.mb_type[mb_xy-1]
2610                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2611                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2612                 : 0;
2613     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2614 }
2615
2616 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2617     H264Context *h = *(void**)arg;
2618     MpegEncContext * const s = &h->s;
2619     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2620
2621     s->mb_skip_run= -1;
2622
2623     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2624                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2625
2626     if( h->pps.cabac ) {
2627         /* realign */
2628         align_get_bits( &s->gb );
2629
2630         /* init cabac */
2631         ff_init_cabac_states( &h->cabac);
2632         ff_init_cabac_decoder( &h->cabac,
2633                                s->gb.buffer + get_bits_count(&s->gb)/8,
2634                                (get_bits_left(&s->gb) + 7)/8);
2635
2636         ff_h264_init_cabac_states(h);
2637
2638         for(;;){
2639 //START_TIMER
2640             int ret = ff_h264_decode_mb_cabac(h);
2641             int eos;
2642 //STOP_TIMER("decode_mb_cabac")
2643
2644             if(ret>=0) ff_h264_hl_decode_mb(h);
2645
2646             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2647                 s->mb_y++;
2648
2649                 ret = ff_h264_decode_mb_cabac(h);
2650
2651                 if(ret>=0) ff_h264_hl_decode_mb(h);
2652                 s->mb_y--;
2653             }
2654             eos = get_cabac_terminate( &h->cabac );
2655
2656             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2657                 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);
2658                 return 0;
2659             }
2660             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2661                 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);
2662                 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);
2663                 return -1;
2664             }
2665
2666             if( ++s->mb_x >= s->mb_width ) {
2667                 s->mb_x = 0;
2668                 loop_filter(h);
2669                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2670                 ++s->mb_y;
2671                 if(FIELD_OR_MBAFF_PICTURE) {
2672                     ++s->mb_y;
2673                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2674                         predict_field_decoding_flag(h);
2675                 }
2676             }
2677
2678             if( eos || s->mb_y >= s->mb_height ) {
2679                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2680                 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);
2681                 return 0;
2682             }
2683         }
2684
2685     } else {
2686         for(;;){
2687             int ret = ff_h264_decode_mb_cavlc(h);
2688
2689             if(ret>=0) ff_h264_hl_decode_mb(h);
2690
2691             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2692                 s->mb_y++;
2693                 ret = ff_h264_decode_mb_cavlc(h);
2694
2695                 if(ret>=0) ff_h264_hl_decode_mb(h);
2696                 s->mb_y--;
2697             }
2698
2699             if(ret<0){
2700                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2701                 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);
2702
2703                 return -1;
2704             }
2705
2706             if(++s->mb_x >= s->mb_width){
2707                 s->mb_x=0;
2708                 loop_filter(h);
2709                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2710                 ++s->mb_y;
2711                 if(FIELD_OR_MBAFF_PICTURE) {
2712                     ++s->mb_y;
2713                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2714                         predict_field_decoding_flag(h);
2715                 }
2716                 if(s->mb_y >= s->mb_height){
2717                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2718
2719                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2720                         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);
2721
2722                         return 0;
2723                     }else{
2724                         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);
2725
2726                         return -1;
2727                     }
2728                 }
2729             }
2730
2731             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2732                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2733                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2734                     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);
2735
2736                     return 0;
2737                 }else{
2738                     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);
2739
2740                     return -1;
2741                 }
2742             }
2743         }
2744     }
2745
2746 #if 0
2747     for(;s->mb_y < s->mb_height; s->mb_y++){
2748         for(;s->mb_x < s->mb_width; s->mb_x++){
2749             int ret= decode_mb(h);
2750
2751             ff_h264_hl_decode_mb(h);
2752
2753             if(ret<0){
2754                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2755                 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);
2756
2757                 return -1;
2758             }
2759
2760             if(++s->mb_x >= s->mb_width){
2761                 s->mb_x=0;
2762                 if(++s->mb_y >= s->mb_height){
2763                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2764                         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);
2765
2766                         return 0;
2767                     }else{
2768                         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);
2769
2770                         return -1;
2771                     }
2772                 }
2773             }
2774
2775             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2776                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2777                     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);
2778
2779                     return 0;
2780                 }else{
2781                     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);
2782
2783                     return -1;
2784                 }
2785             }
2786         }
2787         s->mb_x=0;
2788         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2789     }
2790 #endif
2791     return -1; //not reached
2792 }
2793
2794 /**
2795  * Call decode_slice() for each context.
2796  *
2797  * @param h h264 master context
2798  * @param context_count number of contexts to execute
2799  */
2800 static void execute_decode_slices(H264Context *h, int context_count){
2801     MpegEncContext * const s = &h->s;
2802     AVCodecContext * const avctx= s->avctx;
2803     H264Context *hx;
2804     int i;
2805
2806     if (s->avctx->hwaccel)
2807         return;
2808     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2809         return;
2810     if(context_count == 1) {
2811         decode_slice(avctx, &h);
2812     } else {
2813         for(i = 1; i < context_count; i++) {
2814             hx = h->thread_context[i];
2815             hx->s.error_recognition = avctx->error_recognition;
2816             hx->s.error_count = 0;
2817         }
2818
2819         avctx->execute(avctx, (void *)decode_slice,
2820                        h->thread_context, NULL, context_count, sizeof(void*));
2821
2822         /* pull back stuff from slices to master context */
2823         hx = h->thread_context[context_count - 1];
2824         s->mb_x = hx->s.mb_x;
2825         s->mb_y = hx->s.mb_y;
2826         s->dropable = hx->s.dropable;
2827         s->picture_structure = hx->s.picture_structure;
2828         for(i = 1; i < context_count; i++)
2829             h->s.error_count += h->thread_context[i]->s.error_count;
2830     }
2831 }
2832
2833
2834 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2835     MpegEncContext * const s = &h->s;
2836     AVCodecContext * const avctx= s->avctx;
2837     int buf_index=0;
2838     H264Context *hx; ///< thread context
2839     int context_count = 0;
2840     int next_avc= h->is_avc ? 0 : buf_size;
2841
2842     h->max_contexts = avctx->thread_count;
2843     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2844         h->current_slice = 0;
2845         if (!s->first_field)
2846             s->current_picture_ptr= NULL;
2847         ff_h264_reset_sei(h);
2848     }
2849
2850     for(;;){
2851         int consumed;
2852         int dst_length;
2853         int bit_length;
2854         const uint8_t *ptr;
2855         int i, nalsize = 0;
2856         int err;
2857
2858         if(buf_index >= next_avc) {
2859             if(buf_index >= buf_size) break;
2860             nalsize = 0;
2861             for(i = 0; i < h->nal_length_size; i++)
2862                 nalsize = (nalsize << 8) | buf[buf_index++];
2863             if(nalsize <= 0 || nalsize > buf_size - buf_index){
2864                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2865                 break;
2866             }
2867             next_avc= buf_index + nalsize;
2868         } else {
2869             // start code prefix search
2870             for(; buf_index + 3 < next_avc; buf_index++){
2871                 // This should always succeed in the first iteration.
2872                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2873                     break;
2874             }
2875
2876             if(buf_index+3 >= buf_size) break;
2877
2878             buf_index+=3;
2879             if(buf_index >= next_avc) continue;
2880         }
2881
2882         hx = h->thread_context[context_count];
2883
2884         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2885         if (ptr==NULL || dst_length < 0){
2886             return -1;
2887         }
2888         i= buf_index + consumed;
2889         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2890            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2891             s->workaround_bugs |= FF_BUG_TRUNCATED;
2892
2893         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2894         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2895             dst_length--;
2896         }
2897         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2898
2899         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2900             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);
2901         }
2902
2903         if (h->is_avc && (nalsize != consumed) && nalsize){
2904             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2905         }
2906
2907         buf_index += consumed;
2908
2909         //FIXME do not discard SEI id
2910         if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0)
2911             continue;
2912
2913       again:
2914         err = 0;
2915         switch(hx->nal_unit_type){
2916         case NAL_IDR_SLICE:
2917             if (h->nal_unit_type != NAL_IDR_SLICE) {
2918                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2919                 return -1;
2920             }
2921             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2922         case NAL_SLICE:
2923             init_get_bits(&hx->s.gb, ptr, bit_length);
2924             hx->intra_gb_ptr=
2925             hx->inter_gb_ptr= &hx->s.gb;
2926             hx->s.data_partitioning = 0;
2927
2928             if((err = decode_slice_header(hx, h)))
2929                break;
2930
2931             if (h->current_slice == 1) {
2932                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2933                     return -1;
2934                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2935                     ff_vdpau_h264_picture_start(s);
2936             }
2937
2938             s->current_picture_ptr->key_frame |=
2939                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2940                     (h->sei_recovery_frame_cnt >= 0);
2941             if(hx->redundant_pic_count==0
2942                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2943                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
2944                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
2945                && avctx->skip_frame < AVDISCARD_ALL){
2946                 if(avctx->hwaccel) {
2947                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2948                         return -1;
2949                 }else
2950                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2951                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2952                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2953                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2954                 }else
2955                     context_count++;
2956             }
2957             break;
2958         case NAL_DPA:
2959             init_get_bits(&hx->s.gb, ptr, bit_length);
2960             hx->intra_gb_ptr=
2961             hx->inter_gb_ptr= NULL;
2962
2963             if ((err = decode_slice_header(hx, h)) < 0)
2964                 break;
2965
2966             hx->s.data_partitioning = 1;
2967
2968             break;
2969         case NAL_DPB:
2970             init_get_bits(&hx->intra_gb, ptr, bit_length);
2971             hx->intra_gb_ptr= &hx->intra_gb;
2972             break;
2973         case NAL_DPC:
2974             init_get_bits(&hx->inter_gb, ptr, bit_length);
2975             hx->inter_gb_ptr= &hx->inter_gb;
2976
2977             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2978                && s->context_initialized
2979                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2980                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
2981                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
2982                && avctx->skip_frame < AVDISCARD_ALL)
2983                 context_count++;
2984             break;
2985         case NAL_SEI:
2986             init_get_bits(&s->gb, ptr, bit_length);
2987             ff_h264_decode_sei(h);
2988             break;
2989         case NAL_SPS:
2990             init_get_bits(&s->gb, ptr, bit_length);
2991             ff_h264_decode_seq_parameter_set(h);
2992
2993             if(s->flags& CODEC_FLAG_LOW_DELAY)
2994                 s->low_delay=1;
2995
2996             if(avctx->has_b_frames < 2)
2997                 avctx->has_b_frames= !s->low_delay;
2998             break;
2999         case NAL_PPS:
3000             init_get_bits(&s->gb, ptr, bit_length);
3001
3002             ff_h264_decode_picture_parameter_set(h, bit_length);
3003
3004             break;
3005         case NAL_AUD:
3006         case NAL_END_SEQUENCE:
3007         case NAL_END_STREAM:
3008         case NAL_FILLER_DATA:
3009         case NAL_SPS_EXT:
3010         case NAL_AUXILIARY_SLICE:
3011             break;
3012         default:
3013             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
3014         }
3015
3016         if(context_count == h->max_contexts) {
3017             execute_decode_slices(h, context_count);
3018             context_count = 0;
3019         }
3020
3021         if (err < 0)
3022             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
3023         else if(err == 1) {
3024             /* Slice could not be decoded in parallel mode, copy down
3025              * NAL unit stuff to context 0 and restart. Note that
3026              * rbsp_buffer is not transferred, but since we no longer
3027              * run in parallel mode this should not be an issue. */
3028             h->nal_unit_type = hx->nal_unit_type;
3029             h->nal_ref_idc   = hx->nal_ref_idc;
3030             hx = h;
3031             goto again;
3032         }
3033     }
3034     if(context_count)
3035         execute_decode_slices(h, context_count);
3036     return buf_index;
3037 }
3038
3039 /**
3040  * returns the number of bytes consumed for building the current frame
3041  */
3042 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
3043         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
3044         if(pos+10>buf_size) pos=buf_size; // oops ;)
3045
3046         return pos;
3047 }
3048
3049 static int decode_frame(AVCodecContext *avctx,
3050                              void *data, int *data_size,
3051                              AVPacket *avpkt)
3052 {
3053     const uint8_t *buf = avpkt->data;
3054     int buf_size = avpkt->size;
3055     H264Context *h = avctx->priv_data;
3056     MpegEncContext *s = &h->s;
3057     AVFrame *pict = data;
3058     int buf_index;
3059
3060     s->flags= avctx->flags;
3061     s->flags2= avctx->flags2;
3062
3063    /* end of stream, output what is still in the buffers */
3064  out:
3065     if (buf_size == 0) {
3066         Picture *out;
3067         int i, out_idx;
3068
3069 //FIXME factorize this with the output code below
3070         out = h->delayed_pic[0];
3071         out_idx = 0;
3072         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3073             if(h->delayed_pic[i]->poc < out->poc){
3074                 out = h->delayed_pic[i];
3075                 out_idx = i;
3076             }
3077
3078         for(i=out_idx; h->delayed_pic[i]; i++)
3079             h->delayed_pic[i] = h->delayed_pic[i+1];
3080
3081         if(out){
3082             *data_size = sizeof(AVFrame);
3083             *pict= *(AVFrame*)out;
3084         }
3085
3086         return 0;
3087     }
3088
3089     buf_index=decode_nal_units(h, buf, buf_size);
3090     if(buf_index < 0)
3091         return -1;
3092
3093     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
3094         buf_size = 0;
3095         goto out;
3096     }
3097
3098     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3099         if (avctx->skip_frame >= AVDISCARD_NONREF)
3100             return 0;
3101         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3102         return -1;
3103     }
3104
3105     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3106         Picture *out = s->current_picture_ptr;
3107         Picture *cur = s->current_picture_ptr;
3108         int i, pics, out_of_order, out_idx;
3109
3110         field_end(h);
3111
3112         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
3113             /* Wait for second field. */
3114             *data_size = 0;
3115
3116         } else {
3117             cur->interlaced_frame = 0;
3118             cur->repeat_pict = 0;
3119
3120             /* Signal interlacing information externally. */
3121             /* Prioritize picture timing SEI information over used decoding process if it exists. */
3122
3123             if(h->sps.pic_struct_present_flag){
3124                 switch (h->sei_pic_struct)
3125                 {
3126                 case SEI_PIC_STRUCT_FRAME:
3127                     break;
3128                 case SEI_PIC_STRUCT_TOP_FIELD:
3129                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
3130                     cur->interlaced_frame = 1;
3131                     break;
3132                 case SEI_PIC_STRUCT_TOP_BOTTOM:
3133                 case SEI_PIC_STRUCT_BOTTOM_TOP:
3134                     if (FIELD_OR_MBAFF_PICTURE)
3135                         cur->interlaced_frame = 1;
3136                     else
3137                         // try to flag soft telecine progressive
3138                         cur->interlaced_frame = h->prev_interlaced_frame;
3139                     break;
3140                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
3141                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
3142                     // Signal the possibility of telecined film externally (pic_struct 5,6)
3143                     // From these hints, let the applications decide if they apply deinterlacing.
3144                     cur->repeat_pict = 1;
3145                     break;
3146                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
3147                     // Force progressive here, as doubling interlaced frame is a bad idea.
3148                     cur->repeat_pict = 2;
3149                     break;
3150                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
3151                     cur->repeat_pict = 4;
3152                     break;
3153                 }
3154
3155                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
3156                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
3157             }else{
3158                 /* Derive interlacing flag from used decoding process. */
3159                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
3160             }
3161             h->prev_interlaced_frame = cur->interlaced_frame;
3162
3163             if (cur->field_poc[0] != cur->field_poc[1]){
3164                 /* Derive top_field_first from field pocs. */
3165                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
3166             }else{
3167                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
3168                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
3169                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
3170                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
3171                         cur->top_field_first = 1;
3172                     else
3173                         cur->top_field_first = 0;
3174                 }else{
3175                     /* Most likely progressive */
3176                     cur->top_field_first = 0;
3177                 }
3178             }
3179
3180         //FIXME do something with unavailable reference frames
3181
3182             /* Sort B-frames into display order */
3183
3184             if(h->sps.bitstream_restriction_flag
3185                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
3186                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
3187                 s->low_delay = 0;
3188             }
3189
3190             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
3191                && !h->sps.bitstream_restriction_flag){
3192                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
3193                 s->low_delay= 0;
3194             }
3195
3196             pics = 0;
3197             while(h->delayed_pic[pics]) pics++;
3198
3199             assert(pics <= MAX_DELAYED_PIC_COUNT);
3200
3201             h->delayed_pic[pics++] = cur;
3202             if(cur->reference == 0)
3203                 cur->reference = DELAYED_PIC_REF;
3204
3205             out = h->delayed_pic[0];
3206             out_idx = 0;
3207             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3208                 if(h->delayed_pic[i]->poc < out->poc){
3209                     out = h->delayed_pic[i];
3210                     out_idx = i;
3211                 }
3212             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
3213                 h->outputed_poc= INT_MIN;
3214             out_of_order = out->poc < h->outputed_poc;
3215
3216             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
3217                 { }
3218             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
3219                || (s->low_delay &&
3220                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
3221                  || cur->pict_type == AV_PICTURE_TYPE_B)))
3222             {
3223                 s->low_delay = 0;
3224                 s->avctx->has_b_frames++;
3225             }
3226
3227             if(out_of_order || pics > s->avctx->has_b_frames){
3228                 out->reference &= ~DELAYED_PIC_REF;
3229                 for(i=out_idx; h->delayed_pic[i]; i++)
3230                     h->delayed_pic[i] = h->delayed_pic[i+1];
3231             }
3232             if(!out_of_order && pics > s->avctx->has_b_frames){
3233                 *data_size = sizeof(AVFrame);
3234
3235                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
3236                     h->outputed_poc = INT_MIN;
3237                 } else
3238                     h->outputed_poc = out->poc;
3239                 *pict= *(AVFrame*)out;
3240             }else{
3241                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
3242             }
3243         }
3244     }
3245
3246     assert(pict->data[0] || !*data_size);
3247     ff_print_debug_info(s, pict);
3248 //printf("out %d\n", (int)pict->data[0]);
3249
3250     return get_consumed_bytes(s, buf_index, buf_size);
3251 }
3252 #if 0
3253 static inline void fill_mb_avail(H264Context *h){
3254     MpegEncContext * const s = &h->s;
3255     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3256
3257     if(s->mb_y){
3258         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3259         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
3260         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3261     }else{
3262         h->mb_avail[0]=
3263         h->mb_avail[1]=
3264         h->mb_avail[2]= 0;
3265     }
3266     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3267     h->mb_avail[4]= 1; //FIXME move out
3268     h->mb_avail[5]= 0; //FIXME move out
3269 }
3270 #endif
3271
3272 #ifdef TEST
3273 #undef printf
3274 #undef random
3275 #define COUNT 8000
3276 #define SIZE (COUNT*40)
3277 int main(void){
3278     int i;
3279     uint8_t temp[SIZE];
3280     PutBitContext pb;
3281     GetBitContext gb;
3282 //    int int_temp[10000];
3283     DSPContext dsp;
3284     AVCodecContext avctx;
3285
3286     dsputil_init(&dsp, &avctx);
3287
3288     init_put_bits(&pb, temp, SIZE);
3289     printf("testing unsigned exp golomb\n");
3290     for(i=0; i<COUNT; i++){
3291         START_TIMER
3292         set_ue_golomb(&pb, i);
3293         STOP_TIMER("set_ue_golomb");
3294     }
3295     flush_put_bits(&pb);
3296
3297     init_get_bits(&gb, temp, 8*SIZE);
3298     for(i=0; i<COUNT; i++){
3299         int j, s;
3300
3301         s= show_bits(&gb, 24);
3302
3303         START_TIMER
3304         j= get_ue_golomb(&gb);
3305         if(j != i){
3306             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3307 //            return -1;
3308         }
3309         STOP_TIMER("get_ue_golomb");
3310     }
3311
3312
3313     init_put_bits(&pb, temp, SIZE);
3314     printf("testing signed exp golomb\n");
3315     for(i=0; i<COUNT; i++){
3316         START_TIMER
3317         set_se_golomb(&pb, i - COUNT/2);
3318         STOP_TIMER("set_se_golomb");
3319     }
3320     flush_put_bits(&pb);
3321
3322     init_get_bits(&gb, temp, 8*SIZE);
3323     for(i=0; i<COUNT; i++){
3324         int j, s;
3325
3326         s= show_bits(&gb, 24);
3327
3328         START_TIMER
3329         j= get_se_golomb(&gb);
3330         if(j != i - COUNT/2){
3331             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3332 //            return -1;
3333         }
3334         STOP_TIMER("get_se_golomb");
3335     }
3336
3337 #if 0
3338     printf("testing 4x4 (I)DCT\n");
3339
3340     DCTELEM block[16];
3341     uint8_t src[16], ref[16];
3342     uint64_t error= 0, max_error=0;
3343
3344     for(i=0; i<COUNT; i++){
3345         int j;
3346 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3347         for(j=0; j<16; j++){
3348             ref[j]= random()%255;
3349             src[j]= random()%255;
3350         }
3351
3352         h264_diff_dct_c(block, src, ref, 4);
3353
3354         //normalize
3355         for(j=0; j<16; j++){
3356 //            printf("%d ", block[j]);
3357             block[j]= block[j]*4;
3358             if(j&1) block[j]= (block[j]*4 + 2)/5;
3359             if(j&4) block[j]= (block[j]*4 + 2)/5;
3360         }
3361 //        printf("\n");
3362
3363         h->h264dsp.h264_idct_add(ref, block, 4);
3364 /*        for(j=0; j<16; j++){
3365             printf("%d ", ref[j]);
3366         }
3367         printf("\n");*/
3368
3369         for(j=0; j<16; j++){
3370             int diff= FFABS(src[j] - ref[j]);
3371
3372             error+= diff*diff;
3373             max_error= FFMAX(max_error, diff);
3374         }
3375     }
3376     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3377     printf("testing quantizer\n");
3378     for(qp=0; qp<52; qp++){
3379         for(i=0; i<16; i++)
3380             src1_block[i]= src2_block[i]= random()%255;
3381
3382     }
3383     printf("Testing NAL layer\n");
3384
3385     uint8_t bitstream[COUNT];
3386     uint8_t nal[COUNT*2];
3387     H264Context h;
3388     memset(&h, 0, sizeof(H264Context));
3389
3390     for(i=0; i<COUNT; i++){
3391         int zeros= i;
3392         int nal_length;
3393         int consumed;
3394         int out_length;
3395         uint8_t *out;
3396         int j;
3397
3398         for(j=0; j<COUNT; j++){
3399             bitstream[j]= (random() % 255) + 1;
3400         }
3401
3402         for(j=0; j<zeros; j++){
3403             int pos= random() % COUNT;
3404             while(bitstream[pos] == 0){
3405                 pos++;
3406                 pos %= COUNT;
3407             }
3408             bitstream[pos]=0;
3409         }
3410
3411         START_TIMER
3412
3413         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3414         if(nal_length<0){
3415             printf("encoding failed\n");
3416             return -1;
3417         }
3418
3419         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3420
3421         STOP_TIMER("NAL")
3422
3423         if(out_length != COUNT){
3424             printf("incorrect length %d %d\n", out_length, COUNT);
3425             return -1;
3426         }
3427
3428         if(consumed != nal_length){
3429             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3430             return -1;
3431         }
3432
3433         if(memcmp(bitstream, out, COUNT)){
3434             printf("mismatch\n");
3435             return -1;
3436         }
3437     }
3438 #endif
3439
3440     printf("Testing RBSP\n");
3441
3442
3443     return 0;
3444 }
3445 #endif /* TEST */
3446
3447
3448 av_cold void ff_h264_free_context(H264Context *h)
3449 {
3450     int i;
3451
3452     free_tables(h, 1); //FIXME cleanup init stuff perhaps
3453
3454     for(i = 0; i < MAX_SPS_COUNT; i++)
3455         av_freep(h->sps_buffers + i);
3456
3457     for(i = 0; i < MAX_PPS_COUNT; i++)
3458         av_freep(h->pps_buffers + i);
3459 }
3460
3461 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3462 {
3463     H264Context *h = avctx->priv_data;
3464     MpegEncContext *s = &h->s;
3465
3466     ff_h264_free_context(h);
3467
3468     MPV_common_end(s);
3469
3470 //    memset(h, 0, sizeof(H264Context));
3471
3472     return 0;
3473 }
3474
3475 static const AVProfile profiles[] = {
3476     { FF_PROFILE_H264_BASELINE,             "Baseline"              },
3477     { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline"  },
3478     { FF_PROFILE_H264_MAIN,                 "Main"                  },
3479     { FF_PROFILE_H264_EXTENDED,             "Extended"              },
3480     { FF_PROFILE_H264_HIGH,                 "High"                  },
3481     { FF_PROFILE_H264_HIGH_10,              "High 10"               },
3482     { FF_PROFILE_H264_HIGH_10_INTRA,        "High 10 Intra"         },
3483     { FF_PROFILE_H264_HIGH_422,             "High 4:2:2"            },
3484     { FF_PROFILE_H264_HIGH_422_INTRA,       "High 4:2:2 Intra"      },
3485     { FF_PROFILE_H264_HIGH_444,             "High 4:4:4"            },
3486     { FF_PROFILE_H264_HIGH_444_PREDICTIVE,  "High 4:4:4 Predictive" },
3487     { FF_PROFILE_H264_HIGH_444_INTRA,       "High 4:4:4 Intra"      },
3488     { FF_PROFILE_H264_CAVLC_444,            "CAVLC 4:4:4"           },
3489     { FF_PROFILE_UNKNOWN },
3490 };
3491
3492 AVCodec ff_h264_decoder = {
3493     "h264",
3494     AVMEDIA_TYPE_VIDEO,
3495     CODEC_ID_H264,
3496     sizeof(H264Context),
3497     ff_h264_decode_init,
3498     NULL,
3499     ff_h264_decode_end,
3500     decode_frame,
3501     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
3502         CODEC_CAP_SLICE_THREADS,
3503     .flush= flush_dpb,
3504     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3505     .profiles = NULL_IF_CONFIG_SMALL(profiles),
3506 };
3507
3508 #if CONFIG_H264_VDPAU_DECODER
3509 AVCodec ff_h264_vdpau_decoder = {
3510     "h264_vdpau",
3511     AVMEDIA_TYPE_VIDEO,
3512     CODEC_ID_H264,
3513     sizeof(H264Context),
3514     ff_h264_decode_init,
3515     NULL,
3516     ff_h264_decode_end,
3517     decode_frame,
3518     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3519     .flush= flush_dpb,
3520     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3521     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3522     .profiles = NULL_IF_CONFIG_SMALL(profiles),
3523 };
3524 #endif