H.264: Use 64-/128-bit write-combining macros for copies
[ffmpeg.git] / libavcodec / h264.c
1 /*
2  * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21
22 /**
23  * @file libavcodec/h264.c
24  * H.264 / AVC / MPEG4 part10 codec.
25  * @author Michael Niedermayer <michaelni@gmx.at>
26  */
27
28 #include "internal.h"
29 #include "dsputil.h"
30 #include "avcodec.h"
31 #include "mpegvideo.h"
32 #include "h264.h"
33 #include "h264data.h"
34 #include "h264_mvpred.h"
35 #include "h264_parser.h"
36 #include "golomb.h"
37 #include "mathops.h"
38 #include "rectangle.h"
39 #include "vdpau_internal.h"
40
41 #include "cabac.h"
42
43 //#undef NDEBUG
44 #include <assert.h>
45
46 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
47 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
48
49 static const uint8_t rem6[52]={
50 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,
51 };
52
53 static const uint8_t div6[52]={
54 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,
55 };
56
57 void ff_h264_write_back_intra_pred_mode(H264Context *h){
58     const int mb_xy= h->mb_xy;
59
60     h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
61     h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
62     h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
63     h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
64     h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
65     h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
66     h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
67 }
68
69 /**
70  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
71  */
72 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
73     MpegEncContext * const s = &h->s;
74     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
75     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
76     int i;
77
78     if(!(h->top_samples_available&0x8000)){
79         for(i=0; i<4; i++){
80             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
81             if(status<0){
82                 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);
83                 return -1;
84             } else if(status){
85                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
86             }
87         }
88     }
89
90     if((h->left_samples_available&0x8888)!=0x8888){
91         static const int mask[4]={0x8000,0x2000,0x80,0x20};
92         for(i=0; i<4; i++){
93             if(!(h->left_samples_available&mask[i])){
94                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
95                 if(status<0){
96                     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);
97                     return -1;
98                 } else if(status){
99                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
100                 }
101             }
102         }
103     }
104
105     return 0;
106 } //FIXME cleanup like ff_h264_check_intra_pred_mode
107
108 /**
109  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
110  */
111 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
112     MpegEncContext * const s = &h->s;
113     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
114     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
115
116     if(mode > 6U) {
117         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);
118         return -1;
119     }
120
121     if(!(h->top_samples_available&0x8000)){
122         mode= top[ mode ];
123         if(mode<0){
124             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);
125             return -1;
126         }
127     }
128
129     if((h->left_samples_available&0x8080) != 0x8080){
130         mode= left[ mode ];
131         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
132             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
133         }
134         if(mode<0){
135             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);
136             return -1;
137         }
138     }
139
140     return mode;
141 }
142
143 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
144     int i, si, di;
145     uint8_t *dst;
146     int bufidx;
147
148 //    src[0]&0x80;                //forbidden bit
149     h->nal_ref_idc= src[0]>>5;
150     h->nal_unit_type= src[0]&0x1F;
151
152     src++; length--;
153 #if 0
154     for(i=0; i<length; i++)
155         printf("%2X ", src[i]);
156 #endif
157
158 #if HAVE_FAST_UNALIGNED
159 # if HAVE_FAST_64BIT
160 #   define RS 7
161     for(i=0; i+1<length; i+=9){
162         if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
163 # else
164 #   define RS 3
165     for(i=0; i+1<length; i+=5){
166         if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
167 # endif
168             continue;
169         if(i>0 && !src[i]) i--;
170         while(src[i]) i++;
171 #else
172 #   define RS 0
173     for(i=0; i+1<length; i+=2){
174         if(src[i]) continue;
175         if(i>0 && src[i-1]==0) i--;
176 #endif
177         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
178             if(src[i+2]!=3){
179                 /* startcode, so we must be past the end */
180                 length=i;
181             }
182             break;
183         }
184         i-= RS;
185     }
186
187     if(i>=length-1){ //no escaped 0
188         *dst_length= length;
189         *consumed= length+1; //+1 for the header
190         return src;
191     }
192
193     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
194     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
195     dst= h->rbsp_buffer[bufidx];
196
197     if (dst == NULL){
198         return NULL;
199     }
200
201 //printf("decoding esc\n");
202     memcpy(dst, src, i);
203     si=di=i;
204     while(si+2<length){
205         //remove escapes (very rare 1:2^22)
206         if(src[si+2]>3){
207             dst[di++]= src[si++];
208             dst[di++]= src[si++];
209         }else if(src[si]==0 && src[si+1]==0){
210             if(src[si+2]==3){ //escape
211                 dst[di++]= 0;
212                 dst[di++]= 0;
213                 si+=3;
214                 continue;
215             }else //next start code
216                 goto nsc;
217         }
218
219         dst[di++]= src[si++];
220     }
221     while(si<length)
222         dst[di++]= src[si++];
223 nsc:
224
225     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
226
227     *dst_length= di;
228     *consumed= si + 1;//+1 for the header
229 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
230     return dst;
231 }
232
233 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
234     int v= *src;
235     int r;
236
237     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
238
239     for(r=1; r<9; r++){
240         if(v&1) return r;
241         v>>=1;
242     }
243     return 0;
244 }
245
246 /**
247  * IDCT transforms the 16 dc values and dequantizes them.
248  * @param qp quantization parameter
249  */
250 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
251 #define stride 16
252     int i;
253     int temp[16]; //FIXME check if this is a good idea
254     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
255     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
256
257 //memset(block, 64, 2*256);
258 //return;
259     for(i=0; i<4; i++){
260         const int offset= y_offset[i];
261         const int z0= block[offset+stride*0] + block[offset+stride*4];
262         const int z1= block[offset+stride*0] - block[offset+stride*4];
263         const int z2= block[offset+stride*1] - block[offset+stride*5];
264         const int z3= block[offset+stride*1] + block[offset+stride*5];
265
266         temp[4*i+0]= z0+z3;
267         temp[4*i+1]= z1+z2;
268         temp[4*i+2]= z1-z2;
269         temp[4*i+3]= z0-z3;
270     }
271
272     for(i=0; i<4; i++){
273         const int offset= x_offset[i];
274         const int z0= temp[4*0+i] + temp[4*2+i];
275         const int z1= temp[4*0+i] - temp[4*2+i];
276         const int z2= temp[4*1+i] - temp[4*3+i];
277         const int z3= temp[4*1+i] + temp[4*3+i];
278
279         block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
280         block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
281         block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
282         block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
283     }
284 }
285
286 #if 0
287 /**
288  * DCT transforms the 16 dc values.
289  * @param qp quantization parameter ??? FIXME
290  */
291 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
292 //    const int qmul= dequant_coeff[qp][0];
293     int i;
294     int temp[16]; //FIXME check if this is a good idea
295     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
296     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
297
298     for(i=0; i<4; i++){
299         const int offset= y_offset[i];
300         const int z0= block[offset+stride*0] + block[offset+stride*4];
301         const int z1= block[offset+stride*0] - block[offset+stride*4];
302         const int z2= block[offset+stride*1] - block[offset+stride*5];
303         const int z3= block[offset+stride*1] + block[offset+stride*5];
304
305         temp[4*i+0]= z0+z3;
306         temp[4*i+1]= z1+z2;
307         temp[4*i+2]= z1-z2;
308         temp[4*i+3]= z0-z3;
309     }
310
311     for(i=0; i<4; i++){
312         const int offset= x_offset[i];
313         const int z0= temp[4*0+i] + temp[4*2+i];
314         const int z1= temp[4*0+i] - temp[4*2+i];
315         const int z2= temp[4*1+i] - temp[4*3+i];
316         const int z3= temp[4*1+i] + temp[4*3+i];
317
318         block[stride*0 +offset]= (z0 + z3)>>1;
319         block[stride*2 +offset]= (z1 + z2)>>1;
320         block[stride*8 +offset]= (z1 - z2)>>1;
321         block[stride*10+offset]= (z0 - z3)>>1;
322     }
323 }
324 #endif
325
326 #undef xStride
327 #undef stride
328
329 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
330     const int stride= 16*2;
331     const int xStride= 16;
332     int a,b,c,d,e;
333
334     a= block[stride*0 + xStride*0];
335     b= block[stride*0 + xStride*1];
336     c= block[stride*1 + xStride*0];
337     d= block[stride*1 + xStride*1];
338
339     e= a-b;
340     a= a+b;
341     b= c-d;
342     c= c+d;
343
344     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
345     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
346     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
347     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
348 }
349
350 #if 0
351 static void chroma_dc_dct_c(DCTELEM *block){
352     const int stride= 16*2;
353     const int xStride= 16;
354     int a,b,c,d,e;
355
356     a= block[stride*0 + xStride*0];
357     b= block[stride*0 + xStride*1];
358     c= block[stride*1 + xStride*0];
359     d= block[stride*1 + xStride*1];
360
361     e= a-b;
362     a= a+b;
363     b= c-d;
364     c= c+d;
365
366     block[stride*0 + xStride*0]= (a+c);
367     block[stride*0 + xStride*1]= (e+b);
368     block[stride*1 + xStride*0]= (a-c);
369     block[stride*1 + xStride*1]= (e-b);
370 }
371 #endif
372
373 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
374                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
375                            int src_x_offset, int src_y_offset,
376                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
377     MpegEncContext * const s = &h->s;
378     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
379     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
380     const int luma_xy= (mx&3) + ((my&3)<<2);
381     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
382     uint8_t * src_cb, * src_cr;
383     int extra_width= h->emu_edge_width;
384     int extra_height= h->emu_edge_height;
385     int emu=0;
386     const int full_mx= mx>>2;
387     const int full_my= my>>2;
388     const int pic_width  = 16*s->mb_width;
389     const int pic_height = 16*s->mb_height >> MB_FIELD;
390
391     if(mx&7) extra_width -= 3;
392     if(my&7) extra_height -= 3;
393
394     if(   full_mx < 0-extra_width
395        || full_my < 0-extra_height
396        || full_mx + 16/*FIXME*/ > pic_width + extra_width
397        || full_my + 16/*FIXME*/ > pic_height + extra_height){
398         ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
399             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
400         emu=1;
401     }
402
403     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
404     if(!square){
405         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
406     }
407
408     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
409
410     if(MB_FIELD){
411         // chroma offset when predicting from a field of opposite parity
412         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
413         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
414     }
415     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
416     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
417
418     if(emu){
419         ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
420             src_cb= s->edge_emu_buffer;
421     }
422     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
423
424     if(emu){
425         ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
426             src_cr= s->edge_emu_buffer;
427     }
428     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
429 }
430
431 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
432                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
433                            int x_offset, int y_offset,
434                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
435                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
436                            int list0, int list1){
437     MpegEncContext * const s = &h->s;
438     qpel_mc_func *qpix_op=  qpix_put;
439     h264_chroma_mc_func chroma_op= chroma_put;
440
441     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
442     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
443     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
444     x_offset += 8*s->mb_x;
445     y_offset += 8*(s->mb_y >> MB_FIELD);
446
447     if(list0){
448         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
449         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
450                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
451                            qpix_op, chroma_op);
452
453         qpix_op=  qpix_avg;
454         chroma_op= chroma_avg;
455     }
456
457     if(list1){
458         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
459         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
460                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
461                            qpix_op, chroma_op);
462     }
463 }
464
465 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
466                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
467                            int x_offset, int y_offset,
468                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
469                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
470                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
471                            int list0, int list1){
472     MpegEncContext * const s = &h->s;
473
474     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
475     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
476     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
477     x_offset += 8*s->mb_x;
478     y_offset += 8*(s->mb_y >> MB_FIELD);
479
480     if(list0 && list1){
481         /* don't optimize for luma-only case, since B-frames usually
482          * use implicit weights => chroma too. */
483         uint8_t *tmp_cb = s->obmc_scratchpad;
484         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
485         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
486         int refn0 = h->ref_cache[0][ scan8[n] ];
487         int refn1 = h->ref_cache[1][ scan8[n] ];
488
489         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
490                     dest_y, dest_cb, dest_cr,
491                     x_offset, y_offset, qpix_put, chroma_put);
492         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
493                     tmp_y, tmp_cb, tmp_cr,
494                     x_offset, y_offset, qpix_put, chroma_put);
495
496         if(h->use_weight == 2){
497             int weight0 = h->implicit_weight[refn0][refn1];
498             int weight1 = 64 - weight0;
499             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
500             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
501             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
502         }else{
503             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
504                             h->luma_weight[0][refn0], h->luma_weight[1][refn1],
505                             h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
506             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
507                             h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
508                             h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
509             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
510                             h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
511                             h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
512         }
513     }else{
514         int list = list1 ? 1 : 0;
515         int refn = h->ref_cache[list][ scan8[n] ];
516         Picture *ref= &h->ref_list[list][refn];
517         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
518                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
519                     qpix_put, chroma_put);
520
521         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
522                        h->luma_weight[list][refn], h->luma_offset[list][refn]);
523         if(h->use_weight_chroma){
524             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
525                              h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
526             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
527                              h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
528         }
529     }
530 }
531
532 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
533                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
534                            int x_offset, int y_offset,
535                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
536                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
537                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
538                            int list0, int list1){
539     if((h->use_weight==2 && list0 && list1
540         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
541        || h->use_weight==1)
542         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
543                          x_offset, y_offset, qpix_put, chroma_put,
544                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
545     else
546         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
547                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
548 }
549
550 static inline void prefetch_motion(H264Context *h, int list){
551     /* fetch pixels for estimated mv 4 macroblocks ahead
552      * optimized for 64byte cache lines */
553     MpegEncContext * const s = &h->s;
554     const int refn = h->ref_cache[list][scan8[0]];
555     if(refn >= 0){
556         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
557         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
558         uint8_t **src= h->ref_list[list][refn].data;
559         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
560         s->dsp.prefetch(src[0]+off, s->linesize, 4);
561         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
562         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
563     }
564 }
565
566 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
567                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
568                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
569                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
570     MpegEncContext * const s = &h->s;
571     const int mb_xy= h->mb_xy;
572     const int mb_type= s->current_picture.mb_type[mb_xy];
573
574     assert(IS_INTER(mb_type));
575
576     prefetch_motion(h, 0);
577
578     if(IS_16X16(mb_type)){
579         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
580                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
581                 weight_op, weight_avg,
582                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
583     }else if(IS_16X8(mb_type)){
584         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
585                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
586                 &weight_op[1], &weight_avg[1],
587                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
588         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
589                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
590                 &weight_op[1], &weight_avg[1],
591                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
592     }else if(IS_8X16(mb_type)){
593         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
594                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
595                 &weight_op[2], &weight_avg[2],
596                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
597         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
598                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
599                 &weight_op[2], &weight_avg[2],
600                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
601     }else{
602         int i;
603
604         assert(IS_8X8(mb_type));
605
606         for(i=0; i<4; i++){
607             const int sub_mb_type= h->sub_mb_type[i];
608             const int n= 4*i;
609             int x_offset= (i&1)<<2;
610             int y_offset= (i&2)<<1;
611
612             if(IS_SUB_8X8(sub_mb_type)){
613                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
614                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
615                     &weight_op[3], &weight_avg[3],
616                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
617             }else if(IS_SUB_8X4(sub_mb_type)){
618                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
619                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
620                     &weight_op[4], &weight_avg[4],
621                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
622                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
623                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
624                     &weight_op[4], &weight_avg[4],
625                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
626             }else if(IS_SUB_4X8(sub_mb_type)){
627                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
628                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
629                     &weight_op[5], &weight_avg[5],
630                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
631                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
632                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
633                     &weight_op[5], &weight_avg[5],
634                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
635             }else{
636                 int j;
637                 assert(IS_SUB_4X4(sub_mb_type));
638                 for(j=0; j<4; j++){
639                     int sub_x_offset= x_offset + 2*(j&1);
640                     int sub_y_offset= y_offset +   (j&2);
641                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
642                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
643                         &weight_op[6], &weight_avg[6],
644                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
645                 }
646             }
647         }
648     }
649
650     prefetch_motion(h, 1);
651 }
652
653
654 static void free_tables(H264Context *h){
655     int i;
656     H264Context *hx;
657     av_freep(&h->intra4x4_pred_mode);
658     av_freep(&h->chroma_pred_mode_table);
659     av_freep(&h->cbp_table);
660     av_freep(&h->mvd_table[0]);
661     av_freep(&h->mvd_table[1]);
662     av_freep(&h->direct_table);
663     av_freep(&h->non_zero_count);
664     av_freep(&h->slice_table_base);
665     h->slice_table= NULL;
666     av_freep(&h->list_counts);
667
668     av_freep(&h->mb2b_xy);
669     av_freep(&h->mb2b8_xy);
670
671     for(i = 0; i < MAX_THREADS; i++) {
672         hx = h->thread_context[i];
673         if(!hx) continue;
674         av_freep(&hx->top_borders[1]);
675         av_freep(&hx->top_borders[0]);
676         av_freep(&hx->s.obmc_scratchpad);
677         av_freep(&hx->rbsp_buffer[1]);
678         av_freep(&hx->rbsp_buffer[0]);
679         hx->rbsp_buffer_size[0] = 0;
680         hx->rbsp_buffer_size[1] = 0;
681         if (i) av_freep(&h->thread_context[i]);
682     }
683 }
684
685 static void init_dequant8_coeff_table(H264Context *h){
686     int i,q,x;
687     const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
688     h->dequant8_coeff[0] = h->dequant8_buffer[0];
689     h->dequant8_coeff[1] = h->dequant8_buffer[1];
690
691     for(i=0; i<2; i++ ){
692         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
693             h->dequant8_coeff[1] = h->dequant8_buffer[0];
694             break;
695         }
696
697         for(q=0; q<52; q++){
698             int shift = div6[q];
699             int idx = rem6[q];
700             for(x=0; x<64; x++)
701                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
702                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
703                     h->pps.scaling_matrix8[i][x]) << shift;
704         }
705     }
706 }
707
708 static void init_dequant4_coeff_table(H264Context *h){
709     int i,j,q,x;
710     const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
711     for(i=0; i<6; i++ ){
712         h->dequant4_coeff[i] = h->dequant4_buffer[i];
713         for(j=0; j<i; j++){
714             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
715                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
716                 break;
717             }
718         }
719         if(j<i)
720             continue;
721
722         for(q=0; q<52; q++){
723             int shift = div6[q] + 2;
724             int idx = rem6[q];
725             for(x=0; x<16; x++)
726                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
727                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
728                     h->pps.scaling_matrix4[i][x]) << shift;
729         }
730     }
731 }
732
733 static void init_dequant_tables(H264Context *h){
734     int i,x;
735     init_dequant4_coeff_table(h);
736     if(h->pps.transform_8x8_mode)
737         init_dequant8_coeff_table(h);
738     if(h->sps.transform_bypass){
739         for(i=0; i<6; i++)
740             for(x=0; x<16; x++)
741                 h->dequant4_coeff[i][0][x] = 1<<6;
742         if(h->pps.transform_8x8_mode)
743             for(i=0; i<2; i++)
744                 for(x=0; x<64; x++)
745                     h->dequant8_coeff[i][0][x] = 1<<6;
746     }
747 }
748
749
750 int ff_h264_alloc_tables(H264Context *h){
751     MpegEncContext * const s = &h->s;
752     const int big_mb_num= s->mb_stride * (s->mb_height+1);
753     int x,y;
754
755     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t), fail)
756
757     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
758     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
759     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
760
761     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
762     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t), fail);
763     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t), fail);
764     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 32*big_mb_num * sizeof(uint8_t) , fail);
765     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
766
767     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
768     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
769
770     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
771     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b8_xy , big_mb_num * sizeof(uint32_t), fail);
772     for(y=0; y<s->mb_height; y++){
773         for(x=0; x<s->mb_width; x++){
774             const int mb_xy= x + y*s->mb_stride;
775             const int b_xy = 4*x + 4*y*h->b_stride;
776             const int b8_xy= 2*x + 2*y*h->b8_stride;
777
778             h->mb2b_xy [mb_xy]= b_xy;
779             h->mb2b8_xy[mb_xy]= b8_xy;
780         }
781     }
782
783     s->obmc_scratchpad = NULL;
784
785     if(!h->dequant4_coeff[0])
786         init_dequant_tables(h);
787
788     return 0;
789 fail:
790     free_tables(h);
791     return -1;
792 }
793
794 /**
795  * Mimic alloc_tables(), but for every context thread.
796  */
797 static void clone_tables(H264Context *dst, H264Context *src){
798     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
799     dst->non_zero_count           = src->non_zero_count;
800     dst->slice_table              = src->slice_table;
801     dst->cbp_table                = src->cbp_table;
802     dst->mb2b_xy                  = src->mb2b_xy;
803     dst->mb2b8_xy                 = src->mb2b8_xy;
804     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
805     dst->mvd_table[0]             = src->mvd_table[0];
806     dst->mvd_table[1]             = src->mvd_table[1];
807     dst->direct_table             = src->direct_table;
808     dst->list_counts              = src->list_counts;
809
810     dst->s.obmc_scratchpad = NULL;
811     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
812 }
813
814 /**
815  * Init context
816  * Allocate buffers which are not shared amongst multiple threads.
817  */
818 static int context_init(H264Context *h){
819     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
820     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
821
822     return 0;
823 fail:
824     return -1; // free_tables will clean up for us
825 }
826
827 static av_cold void common_init(H264Context *h){
828     MpegEncContext * const s = &h->s;
829
830     s->width = s->avctx->width;
831     s->height = s->avctx->height;
832     s->codec_id= s->avctx->codec->id;
833
834     ff_h264_pred_init(&h->hpc, s->codec_id);
835
836     h->dequant_coeff_pps= -1;
837     s->unrestricted_mv=1;
838     s->decode=1; //FIXME
839
840     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
841
842     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
843     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
844 }
845
846 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
847     H264Context *h= avctx->priv_data;
848     MpegEncContext * const s = &h->s;
849
850     MPV_decode_defaults(s);
851
852     s->avctx = avctx;
853     common_init(h);
854
855     s->out_format = FMT_H264;
856     s->workaround_bugs= avctx->workaround_bugs;
857
858     // set defaults
859 //    s->decode_mb= ff_h263_decode_mb;
860     s->quarter_sample = 1;
861     if(!avctx->has_b_frames)
862     s->low_delay= 1;
863
864     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
865
866     ff_h264_decode_init_vlc();
867
868     if(avctx->extradata_size > 0 && avctx->extradata &&
869        *(char *)avctx->extradata == 1){
870         h->is_avc = 1;
871         h->got_avcC = 0;
872     } else {
873         h->is_avc = 0;
874     }
875
876     h->thread_context[0] = h;
877     h->outputed_poc = INT_MIN;
878     h->prev_poc_msb= 1<<16;
879     ff_h264_reset_sei(h);
880     if(avctx->codec_id == CODEC_ID_H264){
881         if(avctx->ticks_per_frame == 1){
882             s->avctx->time_base.den *=2;
883         }
884         avctx->ticks_per_frame = 2;
885     }
886     return 0;
887 }
888
889 int ff_h264_frame_start(H264Context *h){
890     MpegEncContext * const s = &h->s;
891     int i;
892
893     if(MPV_frame_start(s, s->avctx) < 0)
894         return -1;
895     ff_er_frame_start(s);
896     /*
897      * MPV_frame_start uses pict_type to derive key_frame.
898      * This is incorrect for H.264; IDR markings must be used.
899      * Zero here; IDR markings per slice in frame or fields are ORed in later.
900      * See decode_nal_units().
901      */
902     s->current_picture_ptr->key_frame= 0;
903     s->current_picture_ptr->mmco_reset= 0;
904
905     assert(s->linesize && s->uvlinesize);
906
907     for(i=0; i<16; i++){
908         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
909         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
910     }
911     for(i=0; i<4; i++){
912         h->block_offset[16+i]=
913         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
914         h->block_offset[24+16+i]=
915         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
916     }
917
918     /* can't be in alloc_tables because linesize isn't known there.
919      * FIXME: redo bipred weight to not require extra buffer? */
920     for(i = 0; i < s->avctx->thread_count; i++)
921         if(!h->thread_context[i]->s.obmc_scratchpad)
922             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
923
924     /* some macroblocks will be accessed before they're available */
925     if(FRAME_MBAFF || s->avctx->thread_count > 1)
926         memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
927
928 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
929
930     // We mark the current picture as non-reference after allocating it, so
931     // that if we break out due to an error it can be released automatically
932     // in the next MPV_frame_start().
933     // SVQ3 as well as most other codecs have only last/next/current and thus
934     // get released even with set reference, besides SVQ3 and others do not
935     // mark frames as reference later "naturally".
936     if(s->codec_id != CODEC_ID_SVQ3)
937         s->current_picture_ptr->reference= 0;
938
939     s->current_picture_ptr->field_poc[0]=
940     s->current_picture_ptr->field_poc[1]= INT_MAX;
941     assert(s->current_picture_ptr->long_ref==0);
942
943     return 0;
944 }
945
946 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){
947     MpegEncContext * const s = &h->s;
948     uint8_t *top_border;
949     int top_idx = 1;
950
951     src_y  -=   linesize;
952     src_cb -= uvlinesize;
953     src_cr -= uvlinesize;
954
955     if(!simple && FRAME_MBAFF){
956         if(s->mb_y&1){
957             if(!MB_MBAFF){
958                 top_border = h->top_borders[0][s->mb_x];
959                 AV_COPY128(top_border, src_y + 15*linesize);
960                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
961                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
962                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
963                 }
964             }
965         }else if(MB_MBAFF){
966             top_idx = 0;
967         }else
968             return;
969     }
970
971     top_border = h->top_borders[top_idx][s->mb_x];
972     // There are two lines saved, the line above the the top macroblock of a pair,
973     // and the line above the bottom macroblock
974     AV_COPY128(top_border, src_y + 16*linesize);
975
976     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
977         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
978         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
979     }
980 }
981
982 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
983     MpegEncContext * const s = &h->s;
984     int temp8, i;
985     uint64_t temp64;
986     int deblock_left;
987     int deblock_top;
988     int mb_xy;
989     int top_idx = 1;
990     uint8_t *top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
991     uint8_t *top_border    = h->top_borders[top_idx][s->mb_x];
992
993     if(!simple && FRAME_MBAFF){
994         if(s->mb_y&1){
995             if(!MB_MBAFF)
996                 return;
997         }else{
998             top_idx = MB_MBAFF ? 0 : 1;
999         }
1000     }
1001
1002     if(h->deblocking_filter == 2) {
1003         mb_xy = h->mb_xy;
1004         deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
1005         deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
1006     } else {
1007         deblock_left = (s->mb_x > 0);
1008         deblock_top =  (s->mb_y > !!MB_FIELD);
1009     }
1010
1011     src_y  -=   linesize + 1;
1012     src_cb -= uvlinesize + 1;
1013     src_cr -= uvlinesize + 1;
1014
1015 #define XCHG(a,b,xchg)\
1016 if (xchg) AV_SWAP64(b,a);\
1017 else      AV_COPY64(b,a);
1018
1019     if(deblock_top){
1020         if(deblock_left){
1021             XCHG(top_border_m1+8, src_y -7, 1);
1022         }
1023         XCHG(top_border+0, src_y +1, xchg);
1024         XCHG(top_border+8, src_y +9, 1);
1025         if(s->mb_x+1 < s->mb_width){
1026             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1027         }
1028     }
1029
1030     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1031         if(deblock_top){
1032             if(deblock_left){
1033                 XCHG(top_border_m1+16, src_cb -7, 1);
1034                 XCHG(top_border_m1+24, src_cr -7, 1);
1035             }
1036             XCHG(top_border+16, src_cb+1, 1);
1037             XCHG(top_border+24, src_cr+1, 1);
1038         }
1039     }
1040 }
1041
1042 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1043     MpegEncContext * const s = &h->s;
1044     const int mb_x= s->mb_x;
1045     const int mb_y= s->mb_y;
1046     const int mb_xy= h->mb_xy;
1047     const int mb_type= s->current_picture.mb_type[mb_xy];
1048     uint8_t  *dest_y, *dest_cb, *dest_cr;
1049     int linesize, uvlinesize /*dct_offset*/;
1050     int i;
1051     int *block_offset = &h->block_offset[0];
1052     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1053     /* is_h264 should always be true if SVQ3 is disabled. */
1054     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1055     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1056     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1057
1058     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1059     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1060     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1061
1062     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1063     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1064
1065     h->list_counts[mb_xy]= h->list_count;
1066
1067     if (!simple && MB_FIELD) {
1068         linesize   = h->mb_linesize   = s->linesize * 2;
1069         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1070         block_offset = &h->block_offset[24];
1071         if(mb_y&1){ //FIXME move out of this function?
1072             dest_y -= s->linesize*15;
1073             dest_cb-= s->uvlinesize*7;
1074             dest_cr-= s->uvlinesize*7;
1075         }
1076         if(FRAME_MBAFF) {
1077             int list;
1078             for(list=0; list<h->list_count; list++){
1079                 if(!USES_LIST(mb_type, list))
1080                     continue;
1081                 if(IS_16X16(mb_type)){
1082                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1083                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1084                 }else{
1085                     for(i=0; i<16; i+=4){
1086                         int ref = h->ref_cache[list][scan8[i]];
1087                         if(ref >= 0)
1088                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1089                     }
1090                 }
1091             }
1092         }
1093     } else {
1094         linesize   = h->mb_linesize   = s->linesize;
1095         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1096 //        dct_offset = s->linesize * 16;
1097     }
1098
1099     if (!simple && IS_INTRA_PCM(mb_type)) {
1100         for (i=0; i<16; i++) {
1101             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1102         }
1103         for (i=0; i<8; i++) {
1104             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1105             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1106         }
1107     } else {
1108         if(IS_INTRA(mb_type)){
1109             if(h->deblocking_filter)
1110                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1111
1112             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1113                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1114                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1115             }
1116
1117             if(IS_INTRA4x4(mb_type)){
1118                 if(simple || !s->encoding){
1119                     if(IS_8x8DCT(mb_type)){
1120                         if(transform_bypass){
1121                             idct_dc_add =
1122                             idct_add    = s->dsp.add_pixels8;
1123                         }else{
1124                             idct_dc_add = s->dsp.h264_idct8_dc_add;
1125                             idct_add    = s->dsp.h264_idct8_add;
1126                         }
1127                         for(i=0; i<16; i+=4){
1128                             uint8_t * const ptr= dest_y + block_offset[i];
1129                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1130                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1131                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1132                             }else{
1133                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1134                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1135                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1136                                 if(nnz){
1137                                     if(nnz == 1 && h->mb[i*16])
1138                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1139                                     else
1140                                         idct_add   (ptr, h->mb + i*16, linesize);
1141                                 }
1142                             }
1143                         }
1144                     }else{
1145                         if(transform_bypass){
1146                             idct_dc_add =
1147                             idct_add    = s->dsp.add_pixels4;
1148                         }else{
1149                             idct_dc_add = s->dsp.h264_idct_dc_add;
1150                             idct_add    = s->dsp.h264_idct_add;
1151                         }
1152                         for(i=0; i<16; i++){
1153                             uint8_t * const ptr= dest_y + block_offset[i];
1154                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1155
1156                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1157                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1158                             }else{
1159                                 uint8_t *topright;
1160                                 int nnz, tr;
1161                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1162                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1163                                     assert(mb_y || linesize <= block_offset[i]);
1164                                     if(!topright_avail){
1165                                         tr= ptr[3 - linesize]*0x01010101;
1166                                         topright= (uint8_t*) &tr;
1167                                     }else
1168                                         topright= ptr + 4 - linesize;
1169                                 }else
1170                                     topright= NULL;
1171
1172                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1173                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1174                                 if(nnz){
1175                                     if(is_h264){
1176                                         if(nnz == 1 && h->mb[i*16])
1177                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1178                                         else
1179                                             idct_add   (ptr, h->mb + i*16, linesize);
1180                                     }else
1181                                         svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1182                                 }
1183                             }
1184                         }
1185                     }
1186                 }
1187             }else{
1188                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1189                 if(is_h264){
1190                     if(!transform_bypass)
1191                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1192                 }else
1193                     svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1194             }
1195             if(h->deblocking_filter)
1196                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1197         }else if(is_h264){
1198             hl_motion(h, dest_y, dest_cb, dest_cr,
1199                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1200                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1201                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1202         }
1203
1204
1205         if(!IS_INTRA4x4(mb_type)){
1206             if(is_h264){
1207                 if(IS_INTRA16x16(mb_type)){
1208                     if(transform_bypass){
1209                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1210                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1211                         }else{
1212                             for(i=0; i<16; i++){
1213                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1214                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1215                             }
1216                         }
1217                     }else{
1218                          s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1219                     }
1220                 }else if(h->cbp&15){
1221                     if(transform_bypass){
1222                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1223                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1224                         for(i=0; i<16; i+=di){
1225                             if(h->non_zero_count_cache[ scan8[i] ]){
1226                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1227                             }
1228                         }
1229                     }else{
1230                         if(IS_8x8DCT(mb_type)){
1231                             s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1232                         }else{
1233                             s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1234                         }
1235                     }
1236                 }
1237             }else{
1238                 for(i=0; i<16; i++){
1239                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1240                         uint8_t * const ptr= dest_y + block_offset[i];
1241                         svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1242                     }
1243                 }
1244             }
1245         }
1246
1247         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1248             uint8_t *dest[2] = {dest_cb, dest_cr};
1249             if(transform_bypass){
1250                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1251                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1252                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1253                 }else{
1254                     idct_add = s->dsp.add_pixels4;
1255                     for(i=16; i<16+8; i++){
1256                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1257                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1258                     }
1259                 }
1260             }else{
1261                 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp[0], h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1262                 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp[1], h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1263                 if(is_h264){
1264                     idct_add = s->dsp.h264_idct_add;
1265                     idct_dc_add = s->dsp.h264_idct_dc_add;
1266                     for(i=16; i<16+8; i++){
1267                         if(h->non_zero_count_cache[ scan8[i] ])
1268                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1269                         else if(h->mb[i*16])
1270                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1271                     }
1272                 }else{
1273                     for(i=16; i<16+8; i++){
1274                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1275                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1276                             svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1277                         }
1278                     }
1279                 }
1280             }
1281         }
1282     }
1283     if(h->cbp || IS_INTRA(mb_type))
1284         s->dsp.clear_blocks(h->mb);
1285
1286     if(h->deblocking_filter && 0) {
1287         backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
1288         fill_filter_caches(h, mb_type); //FIXME don't fill stuff which isn't used by filter_mb
1289         h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
1290         h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
1291         if (!simple && FRAME_MBAFF) {
1292             ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1293         } else {
1294             ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1295         }
1296     }
1297 }
1298
1299 /**
1300  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1301  */
1302 static void hl_decode_mb_simple(H264Context *h){
1303     hl_decode_mb_internal(h, 1);
1304 }
1305
1306 /**
1307  * Process a macroblock; this handles edge cases, such as interlacing.
1308  */
1309 static void av_noinline hl_decode_mb_complex(H264Context *h){
1310     hl_decode_mb_internal(h, 0);
1311 }
1312
1313 void ff_h264_hl_decode_mb(H264Context *h){
1314     MpegEncContext * const s = &h->s;
1315     const int mb_xy= h->mb_xy;
1316     const int mb_type= s->current_picture.mb_type[mb_xy];
1317     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1318
1319     if (is_complex)
1320         hl_decode_mb_complex(h);
1321     else hl_decode_mb_simple(h);
1322 }
1323
1324 static int pred_weight_table(H264Context *h){
1325     MpegEncContext * const s = &h->s;
1326     int list, i;
1327     int luma_def, chroma_def;
1328
1329     h->use_weight= 0;
1330     h->use_weight_chroma= 0;
1331     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1332     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1333     luma_def = 1<<h->luma_log2_weight_denom;
1334     chroma_def = 1<<h->chroma_log2_weight_denom;
1335
1336     for(list=0; list<2; list++){
1337         h->luma_weight_flag[list]   = 0;
1338         h->chroma_weight_flag[list] = 0;
1339         for(i=0; i<h->ref_count[list]; i++){
1340             int luma_weight_flag, chroma_weight_flag;
1341
1342             luma_weight_flag= get_bits1(&s->gb);
1343             if(luma_weight_flag){
1344                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
1345                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
1346                 if(   h->luma_weight[list][i] != luma_def
1347                    || h->luma_offset[list][i] != 0) {
1348                     h->use_weight= 1;
1349                     h->luma_weight_flag[list]= 1;
1350                 }
1351             }else{
1352                 h->luma_weight[list][i]= luma_def;
1353                 h->luma_offset[list][i]= 0;
1354             }
1355
1356             if(CHROMA){
1357                 chroma_weight_flag= get_bits1(&s->gb);
1358                 if(chroma_weight_flag){
1359                     int j;
1360                     for(j=0; j<2; j++){
1361                         h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1362                         h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1363                         if(   h->chroma_weight[list][i][j] != chroma_def
1364                            || h->chroma_offset[list][i][j] != 0) {
1365                             h->use_weight_chroma= 1;
1366                             h->chroma_weight_flag[list]= 1;
1367                         }
1368                     }
1369                 }else{
1370                     int j;
1371                     for(j=0; j<2; j++){
1372                         h->chroma_weight[list][i][j]= chroma_def;
1373                         h->chroma_offset[list][i][j]= 0;
1374                     }
1375                 }
1376             }
1377         }
1378         if(h->slice_type_nos != FF_B_TYPE) break;
1379     }
1380     h->use_weight= h->use_weight || h->use_weight_chroma;
1381     return 0;
1382 }
1383
1384 static void implicit_weight_table(H264Context *h){
1385     MpegEncContext * const s = &h->s;
1386     int ref0, ref1, i;
1387     int cur_poc = s->current_picture_ptr->poc;
1388
1389     for (i = 0; i < 2; i++) {
1390         h->luma_weight_flag[i]   = 0;
1391         h->chroma_weight_flag[i] = 0;
1392     }
1393
1394     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
1395        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1396         h->use_weight= 0;
1397         h->use_weight_chroma= 0;
1398         return;
1399     }
1400
1401     h->use_weight= 2;
1402     h->use_weight_chroma= 2;
1403     h->luma_log2_weight_denom= 5;
1404     h->chroma_log2_weight_denom= 5;
1405
1406     for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1407         int poc0 = h->ref_list[0][ref0].poc;
1408         for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1409             int poc1 = h->ref_list[1][ref1].poc;
1410             int td = av_clip(poc1 - poc0, -128, 127);
1411             if(td){
1412                 int tb = av_clip(cur_poc - poc0, -128, 127);
1413                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1414                 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1415                 if(dist_scale_factor < -64 || dist_scale_factor > 128)
1416                     h->implicit_weight[ref0][ref1] = 32;
1417                 else
1418                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1419             }else
1420                 h->implicit_weight[ref0][ref1] = 32;
1421         }
1422     }
1423 }
1424
1425 /**
1426  * instantaneous decoder refresh.
1427  */
1428 static void idr(H264Context *h){
1429     ff_h264_remove_all_refs(h);
1430     h->prev_frame_num= 0;
1431     h->prev_frame_num_offset= 0;
1432     h->prev_poc_msb=
1433     h->prev_poc_lsb= 0;
1434 }
1435
1436 /* forget old pics after a seek */
1437 static void flush_dpb(AVCodecContext *avctx){
1438     H264Context *h= avctx->priv_data;
1439     int i;
1440     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1441         if(h->delayed_pic[i])
1442             h->delayed_pic[i]->reference= 0;
1443         h->delayed_pic[i]= NULL;
1444     }
1445     h->outputed_poc= INT_MIN;
1446     h->prev_interlaced_frame = 1;
1447     idr(h);
1448     if(h->s.current_picture_ptr)
1449         h->s.current_picture_ptr->reference= 0;
1450     h->s.first_field= 0;
1451     ff_h264_reset_sei(h);
1452     ff_mpeg_flush(avctx);
1453 }
1454
1455 static int init_poc(H264Context *h){
1456     MpegEncContext * const s = &h->s;
1457     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1458     int field_poc[2];
1459     Picture *cur = s->current_picture_ptr;
1460
1461     h->frame_num_offset= h->prev_frame_num_offset;
1462     if(h->frame_num < h->prev_frame_num)
1463         h->frame_num_offset += max_frame_num;
1464
1465     if(h->sps.poc_type==0){
1466         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1467
1468         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1469             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1470         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1471             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1472         else
1473             h->poc_msb = h->prev_poc_msb;
1474 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1475         field_poc[0] =
1476         field_poc[1] = h->poc_msb + h->poc_lsb;
1477         if(s->picture_structure == PICT_FRAME)
1478             field_poc[1] += h->delta_poc_bottom;
1479     }else if(h->sps.poc_type==1){
1480         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1481         int i;
1482
1483         if(h->sps.poc_cycle_length != 0)
1484             abs_frame_num = h->frame_num_offset + h->frame_num;
1485         else
1486             abs_frame_num = 0;
1487
1488         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1489             abs_frame_num--;
1490
1491         expected_delta_per_poc_cycle = 0;
1492         for(i=0; i < h->sps.poc_cycle_length; i++)
1493             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1494
1495         if(abs_frame_num > 0){
1496             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1497             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1498
1499             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1500             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1501                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1502         } else
1503             expectedpoc = 0;
1504
1505         if(h->nal_ref_idc == 0)
1506             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1507
1508         field_poc[0] = expectedpoc + h->delta_poc[0];
1509         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1510
1511         if(s->picture_structure == PICT_FRAME)
1512             field_poc[1] += h->delta_poc[1];
1513     }else{
1514         int poc= 2*(h->frame_num_offset + h->frame_num);
1515
1516         if(!h->nal_ref_idc)
1517             poc--;
1518
1519         field_poc[0]= poc;
1520         field_poc[1]= poc;
1521     }
1522
1523     if(s->picture_structure != PICT_BOTTOM_FIELD)
1524         s->current_picture_ptr->field_poc[0]= field_poc[0];
1525     if(s->picture_structure != PICT_TOP_FIELD)
1526         s->current_picture_ptr->field_poc[1]= field_poc[1];
1527     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1528
1529     return 0;
1530 }
1531
1532
1533 /**
1534  * initialize scan tables
1535  */
1536 static void init_scan_tables(H264Context *h){
1537     MpegEncContext * const s = &h->s;
1538     int i;
1539     if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1540         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1541         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1542     }else{
1543         for(i=0; i<16; i++){
1544 #define T(x) (x>>2) | ((x<<2) & 0xF)
1545             h->zigzag_scan[i] = T(zigzag_scan[i]);
1546             h-> field_scan[i] = T( field_scan[i]);
1547 #undef T
1548         }
1549     }
1550     if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1551         memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1552         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1553         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1554         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1555     }else{
1556         for(i=0; i<64; i++){
1557 #define T(x) (x>>3) | ((x&7)<<3)
1558             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1559             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1560             h->field_scan8x8[i]        = T(field_scan8x8[i]);
1561             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1562 #undef T
1563         }
1564     }
1565     if(h->sps.transform_bypass){ //FIXME same ugly
1566         h->zigzag_scan_q0          = zigzag_scan;
1567         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1568         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1569         h->field_scan_q0           = field_scan;
1570         h->field_scan8x8_q0        = field_scan8x8;
1571         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1572     }else{
1573         h->zigzag_scan_q0          = h->zigzag_scan;
1574         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1575         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1576         h->field_scan_q0           = h->field_scan;
1577         h->field_scan8x8_q0        = h->field_scan8x8;
1578         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1579     }
1580 }
1581
1582 static void field_end(H264Context *h){
1583     MpegEncContext * const s = &h->s;
1584     AVCodecContext * const avctx= s->avctx;
1585     s->mb_y= 0;
1586
1587     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1588     s->current_picture_ptr->pict_type= s->pict_type;
1589
1590     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1591         ff_vdpau_h264_set_reference_frames(s);
1592
1593     if(!s->dropable) {
1594         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1595         h->prev_poc_msb= h->poc_msb;
1596         h->prev_poc_lsb= h->poc_lsb;
1597     }
1598     h->prev_frame_num_offset= h->frame_num_offset;
1599     h->prev_frame_num= h->frame_num;
1600
1601     if (avctx->hwaccel) {
1602         if (avctx->hwaccel->end_frame(avctx) < 0)
1603             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1604     }
1605
1606     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1607         ff_vdpau_h264_picture_complete(s);
1608
1609     /*
1610      * FIXME: Error handling code does not seem to support interlaced
1611      * when slices span multiple rows
1612      * The ff_er_add_slice calls don't work right for bottom
1613      * fields; they cause massive erroneous error concealing
1614      * Error marking covers both fields (top and bottom).
1615      * This causes a mismatched s->error_count
1616      * and a bad error table. Further, the error count goes to
1617      * INT_MAX when called for bottom field, because mb_y is
1618      * past end by one (callers fault) and resync_mb_y != 0
1619      * causes problems for the first MB line, too.
1620      */
1621     if (!FIELD_PICTURE)
1622         ff_er_frame_end(s);
1623
1624     MPV_frame_end(s);
1625
1626     h->current_slice=0;
1627 }
1628
1629 /**
1630  * Replicates H264 "master" context to thread contexts.
1631  */
1632 static void clone_slice(H264Context *dst, H264Context *src)
1633 {
1634     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1635     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1636     dst->s.current_picture      = src->s.current_picture;
1637     dst->s.linesize             = src->s.linesize;
1638     dst->s.uvlinesize           = src->s.uvlinesize;
1639     dst->s.first_field          = src->s.first_field;
1640
1641     dst->prev_poc_msb           = src->prev_poc_msb;
1642     dst->prev_poc_lsb           = src->prev_poc_lsb;
1643     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1644     dst->prev_frame_num         = src->prev_frame_num;
1645     dst->short_ref_count        = src->short_ref_count;
1646
1647     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1648     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1649     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1650     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1651
1652     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1653     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1654 }
1655
1656 /**
1657  * decodes a slice header.
1658  * This will also call MPV_common_init() and frame_start() as needed.
1659  *
1660  * @param h h264context
1661  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1662  *
1663  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1664  */
1665 static int decode_slice_header(H264Context *h, H264Context *h0){
1666     MpegEncContext * const s = &h->s;
1667     MpegEncContext * const s0 = &h0->s;
1668     unsigned int first_mb_in_slice;
1669     unsigned int pps_id;
1670     int num_ref_idx_active_override_flag;
1671     unsigned int slice_type, tmp, i, j;
1672     int default_ref_list_done = 0;
1673     int last_pic_structure;
1674
1675     s->dropable= h->nal_ref_idc == 0;
1676
1677     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1678         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1679         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1680     }else{
1681         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1682         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1683     }
1684
1685     first_mb_in_slice= get_ue_golomb(&s->gb);
1686
1687     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1688         if(h0->current_slice && FIELD_PICTURE){
1689             field_end(h);
1690         }
1691
1692         h0->current_slice = 0;
1693         if (!s0->first_field)
1694             s->current_picture_ptr= NULL;
1695     }
1696
1697     slice_type= get_ue_golomb_31(&s->gb);
1698     if(slice_type > 9){
1699         av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
1700         return -1;
1701     }
1702     if(slice_type > 4){
1703         slice_type -= 5;
1704         h->slice_type_fixed=1;
1705     }else
1706         h->slice_type_fixed=0;
1707
1708     slice_type= golomb_to_pict_type[ slice_type ];
1709     if (slice_type == FF_I_TYPE
1710         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1711         default_ref_list_done = 1;
1712     }
1713     h->slice_type= slice_type;
1714     h->slice_type_nos= slice_type & 3;
1715
1716     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1717     if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
1718         av_log(h->s.avctx, AV_LOG_ERROR,
1719                "B picture before any references, skipping\n");
1720         return -1;
1721     }
1722
1723     pps_id= get_ue_golomb(&s->gb);
1724     if(pps_id>=MAX_PPS_COUNT){
1725         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1726         return -1;
1727     }
1728     if(!h0->pps_buffers[pps_id]) {
1729         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1730         return -1;
1731     }
1732     h->pps= *h0->pps_buffers[pps_id];
1733
1734     if(!h0->sps_buffers[h->pps.sps_id]) {
1735         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1736         return -1;
1737     }
1738     h->sps = *h0->sps_buffers[h->pps.sps_id];
1739
1740     if(h == h0 && h->dequant_coeff_pps != pps_id){
1741         h->dequant_coeff_pps = pps_id;
1742         init_dequant_tables(h);
1743     }
1744
1745     s->mb_width= h->sps.mb_width;
1746     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1747
1748     h->b_stride=  s->mb_width*4;
1749     h->b8_stride= s->mb_width*2;
1750
1751     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1752     if(h->sps.frame_mbs_only_flag)
1753         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1754     else
1755         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1756
1757     if (s->context_initialized
1758         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
1759         if(h != h0)
1760             return -1;   // width / height changed during parallelized decoding
1761         free_tables(h);
1762         flush_dpb(s->avctx);
1763         MPV_common_end(s);
1764     }
1765     if (!s->context_initialized) {
1766         if(h != h0)
1767             return -1;  // we cant (re-)initialize context during parallel decoding
1768
1769         avcodec_set_dimensions(s->avctx, s->width, s->height);
1770         s->avctx->sample_aspect_ratio= h->sps.sar;
1771         if(!s->avctx->sample_aspect_ratio.den)
1772             s->avctx->sample_aspect_ratio.den = 1;
1773
1774         if(h->sps.video_signal_type_present_flag){
1775             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1776             if(h->sps.colour_description_present_flag){
1777                 s->avctx->color_primaries = h->sps.color_primaries;
1778                 s->avctx->color_trc       = h->sps.color_trc;
1779                 s->avctx->colorspace      = h->sps.colorspace;
1780             }
1781         }
1782
1783         if(h->sps.timing_info_present_flag){
1784             s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
1785             if(h->x264_build > 0 && h->x264_build < 44)
1786                 s->avctx->time_base.den *= 2;
1787             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1788                       s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
1789         }
1790         s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1791         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1792
1793         if (MPV_common_init(s) < 0)
1794             return -1;
1795         s->first_field = 0;
1796         h->prev_interlaced_frame = 1;
1797
1798         init_scan_tables(h);
1799         ff_h264_alloc_tables(h);
1800
1801         for(i = 1; i < s->avctx->thread_count; i++) {
1802             H264Context *c;
1803             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1804             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1805             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1806             c->sps = h->sps;
1807             c->pps = h->pps;
1808             init_scan_tables(c);
1809             clone_tables(c, h);
1810         }
1811
1812         for(i = 0; i < s->avctx->thread_count; i++)
1813             if(context_init(h->thread_context[i]) < 0)
1814                 return -1;
1815     }
1816
1817     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1818
1819     h->mb_mbaff = 0;
1820     h->mb_aff_frame = 0;
1821     last_pic_structure = s0->picture_structure;
1822     if(h->sps.frame_mbs_only_flag){
1823         s->picture_structure= PICT_FRAME;
1824     }else{
1825         if(get_bits1(&s->gb)) { //field_pic_flag
1826             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1827         } else {
1828             s->picture_structure= PICT_FRAME;
1829             h->mb_aff_frame = h->sps.mb_aff;
1830         }
1831     }
1832     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1833
1834     if(h0->current_slice == 0){
1835         while(h->frame_num !=  h->prev_frame_num &&
1836               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1837             av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1838             if (ff_h264_frame_start(h) < 0)
1839                 return -1;
1840             h->prev_frame_num++;
1841             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1842             s->current_picture_ptr->frame_num= h->prev_frame_num;
1843             ff_h264_execute_ref_pic_marking(h, NULL, 0);
1844         }
1845
1846         /* See if we have a decoded first field looking for a pair... */
1847         if (s0->first_field) {
1848             assert(s0->current_picture_ptr);
1849             assert(s0->current_picture_ptr->data[0]);
1850             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1851
1852             /* figure out if we have a complementary field pair */
1853             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1854                 /*
1855                  * Previous field is unmatched. Don't display it, but let it
1856                  * remain for reference if marked as such.
1857                  */
1858                 s0->current_picture_ptr = NULL;
1859                 s0->first_field = FIELD_PICTURE;
1860
1861             } else {
1862                 if (h->nal_ref_idc &&
1863                         s0->current_picture_ptr->reference &&
1864                         s0->current_picture_ptr->frame_num != h->frame_num) {
1865                     /*
1866                      * This and previous field were reference, but had
1867                      * different frame_nums. Consider this field first in
1868                      * pair. Throw away previous field except for reference
1869                      * purposes.
1870                      */
1871                     s0->first_field = 1;
1872                     s0->current_picture_ptr = NULL;
1873
1874                 } else {
1875                     /* Second field in complementary pair */
1876                     s0->first_field = 0;
1877                 }
1878             }
1879
1880         } else {
1881             /* Frame or first field in a potentially complementary pair */
1882             assert(!s0->current_picture_ptr);
1883             s0->first_field = FIELD_PICTURE;
1884         }
1885
1886         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1887             s0->first_field = 0;
1888             return -1;
1889         }
1890     }
1891     if(h != h0)
1892         clone_slice(h, h0);
1893
1894     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1895
1896     assert(s->mb_num == s->mb_width * s->mb_height);
1897     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1898        first_mb_in_slice                    >= s->mb_num){
1899         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1900         return -1;
1901     }
1902     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1903     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1904     if (s->picture_structure == PICT_BOTTOM_FIELD)
1905         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1906     assert(s->mb_y < s->mb_height);
1907
1908     if(s->picture_structure==PICT_FRAME){
1909         h->curr_pic_num=   h->frame_num;
1910         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1911     }else{
1912         h->curr_pic_num= 2*h->frame_num + 1;
1913         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1914     }
1915
1916     if(h->nal_unit_type == NAL_IDR_SLICE){
1917         get_ue_golomb(&s->gb); /* idr_pic_id */
1918     }
1919
1920     if(h->sps.poc_type==0){
1921         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1922
1923         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1924             h->delta_poc_bottom= get_se_golomb(&s->gb);
1925         }
1926     }
1927
1928     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1929         h->delta_poc[0]= get_se_golomb(&s->gb);
1930
1931         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1932             h->delta_poc[1]= get_se_golomb(&s->gb);
1933     }
1934
1935     init_poc(h);
1936
1937     if(h->pps.redundant_pic_cnt_present){
1938         h->redundant_pic_count= get_ue_golomb(&s->gb);
1939     }
1940
1941     //set defaults, might be overridden a few lines later
1942     h->ref_count[0]= h->pps.ref_count[0];
1943     h->ref_count[1]= h->pps.ref_count[1];
1944
1945     if(h->slice_type_nos != FF_I_TYPE){
1946         if(h->slice_type_nos == FF_B_TYPE){
1947             h->direct_spatial_mv_pred= get_bits1(&s->gb);
1948         }
1949         num_ref_idx_active_override_flag= get_bits1(&s->gb);
1950
1951         if(num_ref_idx_active_override_flag){
1952             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1953             if(h->slice_type_nos==FF_B_TYPE)
1954                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1955
1956             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1957                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1958                 h->ref_count[0]= h->ref_count[1]= 1;
1959                 return -1;
1960             }
1961         }
1962         if(h->slice_type_nos == FF_B_TYPE)
1963             h->list_count= 2;
1964         else
1965             h->list_count= 1;
1966     }else
1967         h->list_count= 0;
1968
1969     if(!default_ref_list_done){
1970         ff_h264_fill_default_ref_list(h);
1971     }
1972
1973     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
1974         return -1;
1975
1976     if(h->slice_type_nos!=FF_I_TYPE){
1977         s->last_picture_ptr= &h->ref_list[0][0];
1978         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
1979     }
1980     if(h->slice_type_nos==FF_B_TYPE){
1981         s->next_picture_ptr= &h->ref_list[1][0];
1982         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
1983     }
1984
1985     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
1986        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
1987         pred_weight_table(h);
1988     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
1989         implicit_weight_table(h);
1990     else {
1991         h->use_weight = 0;
1992         for (i = 0; i < 2; i++) {
1993             h->luma_weight_flag[i]   = 0;
1994             h->chroma_weight_flag[i] = 0;
1995         }
1996     }
1997
1998     if(h->nal_ref_idc)
1999         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2000
2001     if(FRAME_MBAFF)
2002         ff_h264_fill_mbaff_ref_list(h);
2003
2004     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2005         ff_h264_direct_dist_scale_factor(h);
2006     ff_h264_direct_ref_list_init(h);
2007
2008     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2009         tmp = get_ue_golomb_31(&s->gb);
2010         if(tmp > 2){
2011             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2012             return -1;
2013         }
2014         h->cabac_init_idc= tmp;
2015     }
2016
2017     h->last_qscale_diff = 0;
2018     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2019     if(tmp>51){
2020         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2021         return -1;
2022     }
2023     s->qscale= tmp;
2024     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2025     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2026     //FIXME qscale / qp ... stuff
2027     if(h->slice_type == FF_SP_TYPE){
2028         get_bits1(&s->gb); /* sp_for_switch_flag */
2029     }
2030     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2031         get_se_golomb(&s->gb); /* slice_qs_delta */
2032     }
2033
2034     h->deblocking_filter = 1;
2035     h->slice_alpha_c0_offset = 52;
2036     h->slice_beta_offset = 52;
2037     if( h->pps.deblocking_filter_parameters_present ) {
2038         tmp= get_ue_golomb_31(&s->gb);
2039         if(tmp > 2){
2040             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2041             return -1;
2042         }
2043         h->deblocking_filter= tmp;
2044         if(h->deblocking_filter < 2)
2045             h->deblocking_filter^= 1; // 1<->0
2046
2047         if( h->deblocking_filter ) {
2048             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2049             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2050             if(   h->slice_alpha_c0_offset > 104U
2051                || h->slice_beta_offset     > 104U){
2052                 av_log(s->avctx, AV_LOG_ERROR, "deblocking filter parameters %d %d out of range\n", h->slice_alpha_c0_offset, h->slice_beta_offset);
2053                 return -1;
2054             }
2055         }
2056     }
2057
2058     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2059        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2060        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2061        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2062         h->deblocking_filter= 0;
2063
2064     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2065         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2066             /* Cheat slightly for speed:
2067                Do not bother to deblock across slices. */
2068             h->deblocking_filter = 2;
2069         } else {
2070             h0->max_contexts = 1;
2071             if(!h0->single_decode_warning) {
2072                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2073                 h0->single_decode_warning = 1;
2074             }
2075             if(h != h0)
2076                 return 1; // deblocking switched inside frame
2077         }
2078     }
2079     h->qp_thresh= 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
2080
2081 #if 0 //FMO
2082     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2083         slice_group_change_cycle= get_bits(&s->gb, ?);
2084 #endif
2085
2086     h0->last_slice_type = slice_type;
2087     h->slice_num = ++h0->current_slice;
2088     if(h->slice_num >= MAX_SLICES){
2089         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2090     }
2091
2092     for(j=0; j<2; j++){
2093         int id_list[16];
2094         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2095         for(i=0; i<16; i++){
2096             id_list[i]= 60;
2097             if(h->ref_list[j][i].data[0]){
2098                 int k;
2099                 uint8_t *base= h->ref_list[j][i].base[0];
2100                 for(k=0; k<h->short_ref_count; k++)
2101                     if(h->short_ref[k]->base[0] == base){
2102                         id_list[i]= k;
2103                         break;
2104                     }
2105                 for(k=0; k<h->long_ref_count; k++)
2106                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2107                         id_list[i]= h->short_ref_count + k;
2108                         break;
2109                     }
2110             }
2111         }
2112
2113         ref2frm[0]=
2114         ref2frm[1]= -1;
2115         for(i=0; i<16; i++)
2116             ref2frm[i+2]= 4*id_list[i]
2117                           +(h->ref_list[j][i].reference&3);
2118         ref2frm[18+0]=
2119         ref2frm[18+1]= -1;
2120         for(i=16; i<48; i++)
2121             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2122                           +(h->ref_list[j][i].reference&3);
2123     }
2124
2125     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2126     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2127
2128     s->avctx->refs= h->sps.ref_frame_count;
2129
2130     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2131         av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
2132                h->slice_num,
2133                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2134                first_mb_in_slice,
2135                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2136                pps_id, h->frame_num,
2137                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2138                h->ref_count[0], h->ref_count[1],
2139                s->qscale,
2140                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2141                h->use_weight,
2142                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2143                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2144                );
2145     }
2146
2147     return 0;
2148 }
2149
2150 int ff_h264_get_slice_type(const H264Context *h)
2151 {
2152     switch (h->slice_type) {
2153     case FF_P_TYPE:  return 0;
2154     case FF_B_TYPE:  return 1;
2155     case FF_I_TYPE:  return 2;
2156     case FF_SP_TYPE: return 3;
2157     case FF_SI_TYPE: return 4;
2158     default:         return -1;
2159     }
2160 }
2161
2162 static void loop_filter(H264Context *h){
2163     MpegEncContext * const s = &h->s;
2164     uint8_t  *dest_y, *dest_cb, *dest_cr;
2165     int linesize, uvlinesize, mb_x, mb_y;
2166     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2167     const int old_slice_type= h->slice_type;
2168
2169     if(h->deblocking_filter) {
2170         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2171             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2172                 int list, mb_xy, mb_type;
2173                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2174                 h->slice_num= h->slice_table[mb_xy];
2175                 mb_type= s->current_picture.mb_type[mb_xy];
2176                 h->list_count= h->list_counts[mb_xy];
2177
2178                 if(FRAME_MBAFF)
2179                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2180
2181                 s->mb_x= mb_x;
2182                 s->mb_y= mb_y;
2183                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2184                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2185                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2186                     //FIXME simplify above
2187
2188                 if (MB_FIELD) {
2189                     linesize   = h->mb_linesize   = s->linesize * 2;
2190                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2191                     if(mb_y&1){ //FIXME move out of this function?
2192                         dest_y -= s->linesize*15;
2193                         dest_cb-= s->uvlinesize*7;
2194                         dest_cr-= s->uvlinesize*7;
2195                     }
2196                 } else {
2197                     linesize   = h->mb_linesize   = s->linesize;
2198                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2199                 }
2200                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2201                 if(fill_filter_caches(h, mb_type) < 0)
2202                     continue;
2203                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2204                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2205
2206                 if (FRAME_MBAFF) {
2207                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2208                 } else {
2209                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2210                 }
2211             }
2212         }
2213     }
2214     h->slice_type= old_slice_type;
2215     s->mb_x= 0;
2216     s->mb_y= end_mb_y - FRAME_MBAFF;
2217 }
2218
2219 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2220     H264Context *h = *(void**)arg;
2221     MpegEncContext * const s = &h->s;
2222     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2223
2224     s->mb_skip_run= -1;
2225
2226     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2227                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2228
2229     if( h->pps.cabac ) {
2230         /* realign */
2231         align_get_bits( &s->gb );
2232
2233         /* init cabac */
2234         ff_init_cabac_states( &h->cabac);
2235         ff_init_cabac_decoder( &h->cabac,
2236                                s->gb.buffer + get_bits_count(&s->gb)/8,
2237                                (get_bits_left(&s->gb) + 7)/8);
2238
2239         ff_h264_init_cabac_states(h);
2240
2241         for(;;){
2242 //START_TIMER
2243             int ret = ff_h264_decode_mb_cabac(h);
2244             int eos;
2245 //STOP_TIMER("decode_mb_cabac")
2246
2247             if(ret>=0) ff_h264_hl_decode_mb(h);
2248
2249             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2250                 s->mb_y++;
2251
2252                 ret = ff_h264_decode_mb_cabac(h);
2253
2254                 if(ret>=0) ff_h264_hl_decode_mb(h);
2255                 s->mb_y--;
2256             }
2257             eos = get_cabac_terminate( &h->cabac );
2258
2259             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2260                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%td)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);
2261                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2262                 return -1;
2263             }
2264
2265             if( ++s->mb_x >= s->mb_width ) {
2266                 s->mb_x = 0;
2267                 loop_filter(h);
2268                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2269                 ++s->mb_y;
2270                 if(FIELD_OR_MBAFF_PICTURE) {
2271                     ++s->mb_y;
2272                 }
2273             }
2274
2275             if( eos || s->mb_y >= s->mb_height ) {
2276                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2277                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2278                 return 0;
2279             }
2280         }
2281
2282     } else {
2283         for(;;){
2284             int ret = ff_h264_decode_mb_cavlc(h);
2285
2286             if(ret>=0) ff_h264_hl_decode_mb(h);
2287
2288             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2289                 s->mb_y++;
2290                 ret = ff_h264_decode_mb_cavlc(h);
2291
2292                 if(ret>=0) ff_h264_hl_decode_mb(h);
2293                 s->mb_y--;
2294             }
2295
2296             if(ret<0){
2297                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2298                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2299
2300                 return -1;
2301             }
2302
2303             if(++s->mb_x >= s->mb_width){
2304                 s->mb_x=0;
2305                 loop_filter(h);
2306                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2307                 ++s->mb_y;
2308                 if(FIELD_OR_MBAFF_PICTURE) {
2309                     ++s->mb_y;
2310                 }
2311                 if(s->mb_y >= s->mb_height){
2312                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2313
2314                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2315                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2316
2317                         return 0;
2318                     }else{
2319                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2320
2321                         return -1;
2322                     }
2323                 }
2324             }
2325
2326             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2327                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2328                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2329                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2330
2331                     return 0;
2332                 }else{
2333                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2334
2335                     return -1;
2336                 }
2337             }
2338         }
2339     }
2340
2341 #if 0
2342     for(;s->mb_y < s->mb_height; s->mb_y++){
2343         for(;s->mb_x < s->mb_width; s->mb_x++){
2344             int ret= decode_mb(h);
2345
2346             ff_h264_hl_decode_mb(h);
2347
2348             if(ret<0){
2349                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2350                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2351
2352                 return -1;
2353             }
2354
2355             if(++s->mb_x >= s->mb_width){
2356                 s->mb_x=0;
2357                 if(++s->mb_y >= s->mb_height){
2358                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2359                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2360
2361                         return 0;
2362                     }else{
2363                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2364
2365                         return -1;
2366                     }
2367                 }
2368             }
2369
2370             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2371                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2372                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2373
2374                     return 0;
2375                 }else{
2376                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2377
2378                     return -1;
2379                 }
2380             }
2381         }
2382         s->mb_x=0;
2383         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2384     }
2385 #endif
2386     return -1; //not reached
2387 }
2388
2389 /**
2390  * Call decode_slice() for each context.
2391  *
2392  * @param h h264 master context
2393  * @param context_count number of contexts to execute
2394  */
2395 static void execute_decode_slices(H264Context *h, int context_count){
2396     MpegEncContext * const s = &h->s;
2397     AVCodecContext * const avctx= s->avctx;
2398     H264Context *hx;
2399     int i;
2400
2401     if (s->avctx->hwaccel)
2402         return;
2403     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2404         return;
2405     if(context_count == 1) {
2406         decode_slice(avctx, &h);
2407     } else {
2408         for(i = 1; i < context_count; i++) {
2409             hx = h->thread_context[i];
2410             hx->s.error_recognition = avctx->error_recognition;
2411             hx->s.error_count = 0;
2412         }
2413
2414         avctx->execute(avctx, (void *)decode_slice,
2415                        h->thread_context, NULL, context_count, sizeof(void*));
2416
2417         /* pull back stuff from slices to master context */
2418         hx = h->thread_context[context_count - 1];
2419         s->mb_x = hx->s.mb_x;
2420         s->mb_y = hx->s.mb_y;
2421         s->dropable = hx->s.dropable;
2422         s->picture_structure = hx->s.picture_structure;
2423         for(i = 1; i < context_count; i++)
2424             h->s.error_count += h->thread_context[i]->s.error_count;
2425     }
2426 }
2427
2428
2429 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2430     MpegEncContext * const s = &h->s;
2431     AVCodecContext * const avctx= s->avctx;
2432     int buf_index=0;
2433     H264Context *hx; ///< thread context
2434     int context_count = 0;
2435     int next_avc= h->is_avc ? 0 : buf_size;
2436
2437     h->max_contexts = avctx->thread_count;
2438 #if 0
2439     int i;
2440     for(i=0; i<50; i++){
2441         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2442     }
2443 #endif
2444     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2445         h->current_slice = 0;
2446         if (!s->first_field)
2447             s->current_picture_ptr= NULL;
2448         ff_h264_reset_sei(h);
2449     }
2450
2451     for(;;){
2452         int consumed;
2453         int dst_length;
2454         int bit_length;
2455         const uint8_t *ptr;
2456         int i, nalsize = 0;
2457         int err;
2458
2459         if(buf_index >= next_avc) {
2460             if(buf_index >= buf_size) break;
2461             nalsize = 0;
2462             for(i = 0; i < h->nal_length_size; i++)
2463                 nalsize = (nalsize << 8) | buf[buf_index++];
2464             if(nalsize <= 1 || nalsize > buf_size - buf_index){
2465                 if(nalsize == 1){
2466                     buf_index++;
2467                     continue;
2468                 }else{
2469                     av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2470                     break;
2471                 }
2472             }
2473             next_avc= buf_index + nalsize;
2474         } else {
2475             // start code prefix search
2476             for(; buf_index + 3 < next_avc; buf_index++){
2477                 // This should always succeed in the first iteration.
2478                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2479                     break;
2480             }
2481
2482             if(buf_index+3 >= buf_size) break;
2483
2484             buf_index+=3;
2485             if(buf_index >= next_avc) continue;
2486         }
2487
2488         hx = h->thread_context[context_count];
2489
2490         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2491         if (ptr==NULL || dst_length < 0){
2492             return -1;
2493         }
2494         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2495             dst_length--;
2496         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2497
2498         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2499             av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", hx->nal_unit_type, buf_index, buf_size, dst_length);
2500         }
2501
2502         if (h->is_avc && (nalsize != consumed) && nalsize){
2503             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2504         }
2505
2506         buf_index += consumed;
2507
2508         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2509            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2510             continue;
2511
2512       again:
2513         err = 0;
2514         switch(hx->nal_unit_type){
2515         case NAL_IDR_SLICE:
2516             if (h->nal_unit_type != NAL_IDR_SLICE) {
2517                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2518                 return -1;
2519             }
2520             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2521         case NAL_SLICE:
2522             init_get_bits(&hx->s.gb, ptr, bit_length);
2523             hx->intra_gb_ptr=
2524             hx->inter_gb_ptr= &hx->s.gb;
2525             hx->s.data_partitioning = 0;
2526
2527             if((err = decode_slice_header(hx, h)))
2528                break;
2529
2530             if (s->avctx->hwaccel && h->current_slice == 1) {
2531                 if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2532                     return -1;
2533             }
2534
2535             s->current_picture_ptr->key_frame |=
2536                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2537                     (h->sei_recovery_frame_cnt >= 0);
2538             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2539                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2540                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2541                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2542                && avctx->skip_frame < AVDISCARD_ALL){
2543                 if(avctx->hwaccel) {
2544                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2545                         return -1;
2546                 }else
2547                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2548                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2549                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2550                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2551                 }else
2552                     context_count++;
2553             }
2554             break;
2555         case NAL_DPA:
2556             init_get_bits(&hx->s.gb, ptr, bit_length);
2557             hx->intra_gb_ptr=
2558             hx->inter_gb_ptr= NULL;
2559
2560             if ((err = decode_slice_header(hx, h)) < 0)
2561                 break;
2562
2563             hx->s.data_partitioning = 1;
2564
2565             break;
2566         case NAL_DPB:
2567             init_get_bits(&hx->intra_gb, ptr, bit_length);
2568             hx->intra_gb_ptr= &hx->intra_gb;
2569             break;
2570         case NAL_DPC:
2571             init_get_bits(&hx->inter_gb, ptr, bit_length);
2572             hx->inter_gb_ptr= &hx->inter_gb;
2573
2574             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2575                && s->context_initialized
2576                && s->hurry_up < 5
2577                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2578                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2579                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2580                && avctx->skip_frame < AVDISCARD_ALL)
2581                 context_count++;
2582             break;
2583         case NAL_SEI:
2584             init_get_bits(&s->gb, ptr, bit_length);
2585             ff_h264_decode_sei(h);
2586             break;
2587         case NAL_SPS:
2588             init_get_bits(&s->gb, ptr, bit_length);
2589             ff_h264_decode_seq_parameter_set(h);
2590
2591             if(s->flags& CODEC_FLAG_LOW_DELAY)
2592                 s->low_delay=1;
2593
2594             if(avctx->has_b_frames < 2)
2595                 avctx->has_b_frames= !s->low_delay;
2596             break;
2597         case NAL_PPS:
2598             init_get_bits(&s->gb, ptr, bit_length);
2599
2600             ff_h264_decode_picture_parameter_set(h, bit_length);
2601
2602             break;
2603         case NAL_AUD:
2604         case NAL_END_SEQUENCE:
2605         case NAL_END_STREAM:
2606         case NAL_FILLER_DATA:
2607         case NAL_SPS_EXT:
2608         case NAL_AUXILIARY_SLICE:
2609             break;
2610         default:
2611             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2612         }
2613
2614         if(context_count == h->max_contexts) {
2615             execute_decode_slices(h, context_count);
2616             context_count = 0;
2617         }
2618
2619         if (err < 0)
2620             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2621         else if(err == 1) {
2622             /* Slice could not be decoded in parallel mode, copy down
2623              * NAL unit stuff to context 0 and restart. Note that
2624              * rbsp_buffer is not transferred, but since we no longer
2625              * run in parallel mode this should not be an issue. */
2626             h->nal_unit_type = hx->nal_unit_type;
2627             h->nal_ref_idc   = hx->nal_ref_idc;
2628             hx = h;
2629             goto again;
2630         }
2631     }
2632     if(context_count)
2633         execute_decode_slices(h, context_count);
2634     return buf_index;
2635 }
2636
2637 /**
2638  * returns the number of bytes consumed for building the current frame
2639  */
2640 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2641         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2642         if(pos+10>buf_size) pos=buf_size; // oops ;)
2643
2644         return pos;
2645 }
2646
2647 static int decode_frame(AVCodecContext *avctx,
2648                              void *data, int *data_size,
2649                              AVPacket *avpkt)
2650 {
2651     const uint8_t *buf = avpkt->data;
2652     int buf_size = avpkt->size;
2653     H264Context *h = avctx->priv_data;
2654     MpegEncContext *s = &h->s;
2655     AVFrame *pict = data;
2656     int buf_index;
2657
2658     s->flags= avctx->flags;
2659     s->flags2= avctx->flags2;
2660
2661    /* end of stream, output what is still in the buffers */
2662     if (buf_size == 0) {
2663         Picture *out;
2664         int i, out_idx;
2665
2666 //FIXME factorize this with the output code below
2667         out = h->delayed_pic[0];
2668         out_idx = 0;
2669         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2670             if(h->delayed_pic[i]->poc < out->poc){
2671                 out = h->delayed_pic[i];
2672                 out_idx = i;
2673             }
2674
2675         for(i=out_idx; h->delayed_pic[i]; i++)
2676             h->delayed_pic[i] = h->delayed_pic[i+1];
2677
2678         if(out){
2679             *data_size = sizeof(AVFrame);
2680             *pict= *(AVFrame*)out;
2681         }
2682
2683         return 0;
2684     }
2685
2686     if(h->is_avc && !h->got_avcC) {
2687         int i, cnt, nalsize;
2688         unsigned char *p = avctx->extradata;
2689         if(avctx->extradata_size < 7) {
2690             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
2691             return -1;
2692         }
2693         if(*p != 1) {
2694             av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
2695             return -1;
2696         }
2697         /* sps and pps in the avcC always have length coded with 2 bytes,
2698            so put a fake nal_length_size = 2 while parsing them */
2699         h->nal_length_size = 2;
2700         // Decode sps from avcC
2701         cnt = *(p+5) & 0x1f; // Number of sps
2702         p += 6;
2703         for (i = 0; i < cnt; i++) {
2704             nalsize = AV_RB16(p) + 2;
2705             if(decode_nal_units(h, p, nalsize) < 0) {
2706                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
2707                 return -1;
2708             }
2709             p += nalsize;
2710         }
2711         // Decode pps from avcC
2712         cnt = *(p++); // Number of pps
2713         for (i = 0; i < cnt; i++) {
2714             nalsize = AV_RB16(p) + 2;
2715             if(decode_nal_units(h, p, nalsize)  != nalsize) {
2716                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
2717                 return -1;
2718             }
2719             p += nalsize;
2720         }
2721         // Now store right nal length size, that will be use to parse all other nals
2722         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
2723         // Do not reparse avcC
2724         h->got_avcC = 1;
2725     }
2726
2727     if(!h->got_avcC && !h->is_avc && s->avctx->extradata_size){
2728         if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
2729             return -1;
2730         h->got_avcC = 1;
2731     }
2732
2733     buf_index=decode_nal_units(h, buf, buf_size);
2734     if(buf_index < 0)
2735         return -1;
2736
2737     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2738         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2739         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2740         return -1;
2741     }
2742
2743     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2744         Picture *out = s->current_picture_ptr;
2745         Picture *cur = s->current_picture_ptr;
2746         int i, pics, out_of_order, out_idx;
2747
2748         field_end(h);
2749
2750         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2751             /* Wait for second field. */
2752             *data_size = 0;
2753
2754         } else {
2755             cur->interlaced_frame = 0;
2756             cur->repeat_pict = 0;
2757
2758             /* Signal interlacing information externally. */
2759             /* Prioritize picture timing SEI information over used decoding process if it exists. */
2760
2761             if(h->sps.pic_struct_present_flag){
2762                 switch (h->sei_pic_struct)
2763                 {
2764                 case SEI_PIC_STRUCT_FRAME:
2765                     break;
2766                 case SEI_PIC_STRUCT_TOP_FIELD:
2767                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
2768                     cur->interlaced_frame = 1;
2769                     break;
2770                 case SEI_PIC_STRUCT_TOP_BOTTOM:
2771                 case SEI_PIC_STRUCT_BOTTOM_TOP:
2772                     if (FIELD_OR_MBAFF_PICTURE)
2773                         cur->interlaced_frame = 1;
2774                     else
2775                         // try to flag soft telecine progressive
2776                         cur->interlaced_frame = h->prev_interlaced_frame;
2777                     break;
2778                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2779                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2780                     // Signal the possibility of telecined film externally (pic_struct 5,6)
2781                     // From these hints, let the applications decide if they apply deinterlacing.
2782                     cur->repeat_pict = 1;
2783                     break;
2784                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
2785                     // Force progressive here, as doubling interlaced frame is a bad idea.
2786                     cur->repeat_pict = 2;
2787                     break;
2788                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
2789                     cur->repeat_pict = 4;
2790                     break;
2791                 }
2792
2793                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2794                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
2795             }else{
2796                 /* Derive interlacing flag from used decoding process. */
2797                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
2798             }
2799             h->prev_interlaced_frame = cur->interlaced_frame;
2800
2801             if (cur->field_poc[0] != cur->field_poc[1]){
2802                 /* Derive top_field_first from field pocs. */
2803                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
2804             }else{
2805                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
2806                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
2807                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
2808                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2809                         cur->top_field_first = 1;
2810                     else
2811                         cur->top_field_first = 0;
2812                 }else{
2813                     /* Most likely progressive */
2814                     cur->top_field_first = 0;
2815                 }
2816             }
2817
2818         //FIXME do something with unavailable reference frames
2819
2820             /* Sort B-frames into display order */
2821
2822             if(h->sps.bitstream_restriction_flag
2823                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
2824                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
2825                 s->low_delay = 0;
2826             }
2827
2828             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
2829                && !h->sps.bitstream_restriction_flag){
2830                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
2831                 s->low_delay= 0;
2832             }
2833
2834             pics = 0;
2835             while(h->delayed_pic[pics]) pics++;
2836
2837             assert(pics <= MAX_DELAYED_PIC_COUNT);
2838
2839             h->delayed_pic[pics++] = cur;
2840             if(cur->reference == 0)
2841                 cur->reference = DELAYED_PIC_REF;
2842
2843             out = h->delayed_pic[0];
2844             out_idx = 0;
2845             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2846                 if(h->delayed_pic[i]->poc < out->poc){
2847                     out = h->delayed_pic[i];
2848                     out_idx = i;
2849                 }
2850             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
2851                 h->outputed_poc= INT_MIN;
2852             out_of_order = out->poc < h->outputed_poc;
2853
2854             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
2855                 { }
2856             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
2857                || (s->low_delay &&
2858                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
2859                  || cur->pict_type == FF_B_TYPE)))
2860             {
2861                 s->low_delay = 0;
2862                 s->avctx->has_b_frames++;
2863             }
2864
2865             if(out_of_order || pics > s->avctx->has_b_frames){
2866                 out->reference &= ~DELAYED_PIC_REF;
2867                 for(i=out_idx; h->delayed_pic[i]; i++)
2868                     h->delayed_pic[i] = h->delayed_pic[i+1];
2869             }
2870             if(!out_of_order && pics > s->avctx->has_b_frames){
2871                 *data_size = sizeof(AVFrame);
2872
2873                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
2874                     h->outputed_poc = INT_MIN;
2875                 } else
2876                     h->outputed_poc = out->poc;
2877                 *pict= *(AVFrame*)out;
2878             }else{
2879                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
2880             }
2881         }
2882     }
2883
2884     assert(pict->data[0] || !*data_size);
2885     ff_print_debug_info(s, pict);
2886 //printf("out %d\n", (int)pict->data[0]);
2887
2888     return get_consumed_bytes(s, buf_index, buf_size);
2889 }
2890 #if 0
2891 static inline void fill_mb_avail(H264Context *h){
2892     MpegEncContext * const s = &h->s;
2893     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2894
2895     if(s->mb_y){
2896         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
2897         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
2898         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
2899     }else{
2900         h->mb_avail[0]=
2901         h->mb_avail[1]=
2902         h->mb_avail[2]= 0;
2903     }
2904     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
2905     h->mb_avail[4]= 1; //FIXME move out
2906     h->mb_avail[5]= 0; //FIXME move out
2907 }
2908 #endif
2909
2910 #ifdef TEST
2911 #undef printf
2912 #undef random
2913 #define COUNT 8000
2914 #define SIZE (COUNT*40)
2915 int main(void){
2916     int i;
2917     uint8_t temp[SIZE];
2918     PutBitContext pb;
2919     GetBitContext gb;
2920 //    int int_temp[10000];
2921     DSPContext dsp;
2922     AVCodecContext avctx;
2923
2924     dsputil_init(&dsp, &avctx);
2925
2926     init_put_bits(&pb, temp, SIZE);
2927     printf("testing unsigned exp golomb\n");
2928     for(i=0; i<COUNT; i++){
2929         START_TIMER
2930         set_ue_golomb(&pb, i);
2931         STOP_TIMER("set_ue_golomb");
2932     }
2933     flush_put_bits(&pb);
2934
2935     init_get_bits(&gb, temp, 8*SIZE);
2936     for(i=0; i<COUNT; i++){
2937         int j, s;
2938
2939         s= show_bits(&gb, 24);
2940
2941         START_TIMER
2942         j= get_ue_golomb(&gb);
2943         if(j != i){
2944             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2945 //            return -1;
2946         }
2947         STOP_TIMER("get_ue_golomb");
2948     }
2949
2950
2951     init_put_bits(&pb, temp, SIZE);
2952     printf("testing signed exp golomb\n");
2953     for(i=0; i<COUNT; i++){
2954         START_TIMER
2955         set_se_golomb(&pb, i - COUNT/2);
2956         STOP_TIMER("set_se_golomb");
2957     }
2958     flush_put_bits(&pb);
2959
2960     init_get_bits(&gb, temp, 8*SIZE);
2961     for(i=0; i<COUNT; i++){
2962         int j, s;
2963
2964         s= show_bits(&gb, 24);
2965
2966         START_TIMER
2967         j= get_se_golomb(&gb);
2968         if(j != i - COUNT/2){
2969             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2970 //            return -1;
2971         }
2972         STOP_TIMER("get_se_golomb");
2973     }
2974
2975 #if 0
2976     printf("testing 4x4 (I)DCT\n");
2977
2978     DCTELEM block[16];
2979     uint8_t src[16], ref[16];
2980     uint64_t error= 0, max_error=0;
2981
2982     for(i=0; i<COUNT; i++){
2983         int j;
2984 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
2985         for(j=0; j<16; j++){
2986             ref[j]= random()%255;
2987             src[j]= random()%255;
2988         }
2989
2990         h264_diff_dct_c(block, src, ref, 4);
2991
2992         //normalize
2993         for(j=0; j<16; j++){
2994 //            printf("%d ", block[j]);
2995             block[j]= block[j]*4;
2996             if(j&1) block[j]= (block[j]*4 + 2)/5;
2997             if(j&4) block[j]= (block[j]*4 + 2)/5;
2998         }
2999 //        printf("\n");
3000
3001         s->dsp.h264_idct_add(ref, block, 4);
3002 /*        for(j=0; j<16; j++){
3003             printf("%d ", ref[j]);
3004         }
3005         printf("\n");*/
3006
3007         for(j=0; j<16; j++){
3008             int diff= FFABS(src[j] - ref[j]);
3009
3010             error+= diff*diff;
3011             max_error= FFMAX(max_error, diff);
3012         }
3013     }
3014     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3015     printf("testing quantizer\n");
3016     for(qp=0; qp<52; qp++){
3017         for(i=0; i<16; i++)
3018             src1_block[i]= src2_block[i]= random()%255;
3019
3020     }
3021     printf("Testing NAL layer\n");
3022
3023     uint8_t bitstream[COUNT];
3024     uint8_t nal[COUNT*2];
3025     H264Context h;
3026     memset(&h, 0, sizeof(H264Context));
3027
3028     for(i=0; i<COUNT; i++){
3029         int zeros= i;
3030         int nal_length;
3031         int consumed;
3032         int out_length;
3033         uint8_t *out;
3034         int j;
3035
3036         for(j=0; j<COUNT; j++){
3037             bitstream[j]= (random() % 255) + 1;
3038         }
3039
3040         for(j=0; j<zeros; j++){
3041             int pos= random() % COUNT;
3042             while(bitstream[pos] == 0){
3043                 pos++;
3044                 pos %= COUNT;
3045             }
3046             bitstream[pos]=0;
3047         }
3048
3049         START_TIMER
3050
3051         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3052         if(nal_length<0){
3053             printf("encoding failed\n");
3054             return -1;
3055         }
3056
3057         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3058
3059         STOP_TIMER("NAL")
3060
3061         if(out_length != COUNT){
3062             printf("incorrect length %d %d\n", out_length, COUNT);
3063             return -1;
3064         }
3065
3066         if(consumed != nal_length){
3067             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3068             return -1;
3069         }
3070
3071         if(memcmp(bitstream, out, COUNT)){
3072             printf("mismatch\n");
3073             return -1;
3074         }
3075     }
3076 #endif
3077
3078     printf("Testing RBSP\n");
3079
3080
3081     return 0;
3082 }
3083 #endif /* TEST */
3084
3085
3086 av_cold void ff_h264_free_context(H264Context *h)
3087 {
3088     int i;
3089
3090     free_tables(h); //FIXME cleanup init stuff perhaps
3091
3092     for(i = 0; i < MAX_SPS_COUNT; i++)
3093         av_freep(h->sps_buffers + i);
3094
3095     for(i = 0; i < MAX_PPS_COUNT; i++)
3096         av_freep(h->pps_buffers + i);
3097 }
3098
3099 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3100 {
3101     H264Context *h = avctx->priv_data;
3102     MpegEncContext *s = &h->s;
3103
3104     ff_h264_free_context(h);
3105
3106     MPV_common_end(s);
3107
3108 //    memset(h, 0, sizeof(H264Context));
3109
3110     return 0;
3111 }
3112
3113
3114 AVCodec h264_decoder = {
3115     "h264",
3116     CODEC_TYPE_VIDEO,
3117     CODEC_ID_H264,
3118     sizeof(H264Context),
3119     ff_h264_decode_init,
3120     NULL,
3121     ff_h264_decode_end,
3122     decode_frame,
3123     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3124     .flush= flush_dpb,
3125     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3126     .pix_fmts= ff_hwaccel_pixfmt_list_420,
3127 };
3128
3129 #if CONFIG_H264_VDPAU_DECODER
3130 AVCodec h264_vdpau_decoder = {
3131     "h264_vdpau",
3132     CODEC_TYPE_VIDEO,
3133     CODEC_ID_H264,
3134     sizeof(H264Context),
3135     ff_h264_decode_init,
3136     NULL,
3137     ff_h264_decode_end,
3138     decode_frame,
3139     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3140     .flush= flush_dpb,
3141     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3142     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3143 };
3144 #endif