Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 50c21814

History | View | Annotate | Download (295 KB)

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 h264.c
24
 * H.264 / AVC / MPEG4 part10 codec.
25
 * @author Michael Niedermayer <michaelni@gmx.at>
26
 */
27

    
28
#include "dsputil.h"
29
#include "avcodec.h"
30
#include "mpegvideo.h"
31
#include "h264.h"
32
#include "h264data.h"
33
#include "h264_parser.h"
34
#include "golomb.h"
35

    
36
#include "cabac.h"
37

    
38
//#undef NDEBUG
39
#include <assert.h>
40

    
41
static VLC coeff_token_vlc[4];
42
static VLC chroma_dc_coeff_token_vlc;
43

    
44
static VLC total_zeros_vlc[15];
45
static VLC chroma_dc_total_zeros_vlc[3];
46

    
47
static VLC run_vlc[6];
48
static VLC run7_vlc;
49

    
50
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
51
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
52
static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
53
static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
54

    
55
static av_always_inline uint32_t pack16to32(int a, int b){
56
#ifdef WORDS_BIGENDIAN
57
   return (b&0xFFFF) + (a<<16);
58
#else
59
   return (a&0xFFFF) + (b<<16);
60
#endif
61
}
62

    
63
const uint8_t ff_rem6[52]={
64
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,
65
};
66

    
67
const uint8_t ff_div6[52]={
68
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,
69
};
70

    
71

    
72
/**
73
 * fill a rectangle.
74
 * @param h height of the rectangle, should be a constant
75
 * @param w width of the rectangle, should be a constant
76
 * @param size the size of val (1 or 4), should be a constant
77
 */
78
static av_always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
79
    uint8_t *p= (uint8_t*)vp;
80
    assert(size==1 || size==4);
81
    assert(w<=4);
82

    
83
    w      *= size;
84
    stride *= size;
85

    
86
    assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
87
    assert((stride&(w-1))==0);
88
    if(w==2){
89
        const uint16_t v= size==4 ? val : val*0x0101;
90
        *(uint16_t*)(p + 0*stride)= v;
91
        if(h==1) return;
92
        *(uint16_t*)(p + 1*stride)= v;
93
        if(h==2) return;
94
        *(uint16_t*)(p + 2*stride)= v;
95
        *(uint16_t*)(p + 3*stride)= v;
96
    }else if(w==4){
97
        const uint32_t v= size==4 ? val : val*0x01010101;
98
        *(uint32_t*)(p + 0*stride)= v;
99
        if(h==1) return;
100
        *(uint32_t*)(p + 1*stride)= v;
101
        if(h==2) return;
102
        *(uint32_t*)(p + 2*stride)= v;
103
        *(uint32_t*)(p + 3*stride)= v;
104
    }else if(w==8){
105
    //gcc can't optimize 64bit math on x86_32
106
#if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64)
107
        const uint64_t v= val*0x0100000001ULL;
108
        *(uint64_t*)(p + 0*stride)= v;
109
        if(h==1) return;
110
        *(uint64_t*)(p + 1*stride)= v;
111
        if(h==2) return;
112
        *(uint64_t*)(p + 2*stride)= v;
113
        *(uint64_t*)(p + 3*stride)= v;
114
    }else if(w==16){
115
        const uint64_t v= val*0x0100000001ULL;
116
        *(uint64_t*)(p + 0+0*stride)= v;
117
        *(uint64_t*)(p + 8+0*stride)= v;
118
        *(uint64_t*)(p + 0+1*stride)= v;
119
        *(uint64_t*)(p + 8+1*stride)= v;
120
        if(h==2) return;
121
        *(uint64_t*)(p + 0+2*stride)= v;
122
        *(uint64_t*)(p + 8+2*stride)= v;
123
        *(uint64_t*)(p + 0+3*stride)= v;
124
        *(uint64_t*)(p + 8+3*stride)= v;
125
#else
126
        *(uint32_t*)(p + 0+0*stride)= val;
127
        *(uint32_t*)(p + 4+0*stride)= val;
128
        if(h==1) return;
129
        *(uint32_t*)(p + 0+1*stride)= val;
130
        *(uint32_t*)(p + 4+1*stride)= val;
131
        if(h==2) return;
132
        *(uint32_t*)(p + 0+2*stride)= val;
133
        *(uint32_t*)(p + 4+2*stride)= val;
134
        *(uint32_t*)(p + 0+3*stride)= val;
135
        *(uint32_t*)(p + 4+3*stride)= val;
136
    }else if(w==16){
137
        *(uint32_t*)(p + 0+0*stride)= val;
138
        *(uint32_t*)(p + 4+0*stride)= val;
139
        *(uint32_t*)(p + 8+0*stride)= val;
140
        *(uint32_t*)(p +12+0*stride)= val;
141
        *(uint32_t*)(p + 0+1*stride)= val;
142
        *(uint32_t*)(p + 4+1*stride)= val;
143
        *(uint32_t*)(p + 8+1*stride)= val;
144
        *(uint32_t*)(p +12+1*stride)= val;
145
        if(h==2) return;
146
        *(uint32_t*)(p + 0+2*stride)= val;
147
        *(uint32_t*)(p + 4+2*stride)= val;
148
        *(uint32_t*)(p + 8+2*stride)= val;
149
        *(uint32_t*)(p +12+2*stride)= val;
150
        *(uint32_t*)(p + 0+3*stride)= val;
151
        *(uint32_t*)(p + 4+3*stride)= val;
152
        *(uint32_t*)(p + 8+3*stride)= val;
153
        *(uint32_t*)(p +12+3*stride)= val;
154
#endif
155
    }else
156
        assert(0);
157
    assert(h==4);
158
}
159

    
160
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
161
    MpegEncContext * const s = &h->s;
162
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
163
    int topleft_xy, top_xy, topright_xy, left_xy[2];
164
    int topleft_type, top_type, topright_type, left_type[2];
165
    int left_block[8];
166
    int i;
167

    
168
    //FIXME deblocking could skip the intra and nnz parts.
169
    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[mb_xy-s->mb_stride]) && !FRAME_MBAFF)
170
        return;
171

    
172
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
173

    
174
    top_xy     = mb_xy  - s->mb_stride;
175
    topleft_xy = top_xy - 1;
176
    topright_xy= top_xy + 1;
177
    left_xy[1] = left_xy[0] = mb_xy-1;
178
    left_block[0]= 0;
179
    left_block[1]= 1;
180
    left_block[2]= 2;
181
    left_block[3]= 3;
182
    left_block[4]= 7;
183
    left_block[5]= 10;
184
    left_block[6]= 8;
185
    left_block[7]= 11;
186
    if(FRAME_MBAFF){
187
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
188
        const int top_pair_xy      = pair_xy     - s->mb_stride;
189
        const int topleft_pair_xy  = top_pair_xy - 1;
190
        const int topright_pair_xy = top_pair_xy + 1;
191
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
192
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
193
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
194
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
195
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
196
        const int bottom = (s->mb_y & 1);
197
        tprintf(s->avctx, "fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
198
        if (bottom
199
                ? !curr_mb_frame_flag // bottom macroblock
200
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
201
                ) {
202
            top_xy -= s->mb_stride;
203
        }
204
        if (bottom
205
                ? !curr_mb_frame_flag // bottom macroblock
206
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
207
                ) {
208
            topleft_xy -= s->mb_stride;
209
        }
210
        if (bottom
211
                ? !curr_mb_frame_flag // bottom macroblock
212
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
213
                ) {
214
            topright_xy -= s->mb_stride;
215
        }
216
        if (left_mb_frame_flag != curr_mb_frame_flag) {
217
            left_xy[1] = left_xy[0] = pair_xy - 1;
218
            if (curr_mb_frame_flag) {
219
                if (bottom) {
220
                    left_block[0]= 2;
221
                    left_block[1]= 2;
222
                    left_block[2]= 3;
223
                    left_block[3]= 3;
224
                    left_block[4]= 8;
225
                    left_block[5]= 11;
226
                    left_block[6]= 8;
227
                    left_block[7]= 11;
228
                } else {
229
                    left_block[0]= 0;
230
                    left_block[1]= 0;
231
                    left_block[2]= 1;
232
                    left_block[3]= 1;
233
                    left_block[4]= 7;
234
                    left_block[5]= 10;
235
                    left_block[6]= 7;
236
                    left_block[7]= 10;
237
                }
238
            } else {
239
                left_xy[1] += s->mb_stride;
240
                //left_block[0]= 0;
241
                left_block[1]= 2;
242
                left_block[2]= 0;
243
                left_block[3]= 2;
244
                //left_block[4]= 7;
245
                left_block[5]= 10;
246
                left_block[6]= 7;
247
                left_block[7]= 10;
248
            }
249
        }
250
    }
251

    
252
    h->top_mb_xy = top_xy;
253
    h->left_mb_xy[0] = left_xy[0];
254
    h->left_mb_xy[1] = left_xy[1];
255
    if(for_deblock){
256
        topleft_type = 0;
257
        topright_type = 0;
258
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
259
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
260
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
261

    
262
        if(FRAME_MBAFF && !IS_INTRA(mb_type)){
263
            int list;
264
            int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
265
            for(i=0; i<16; i++)
266
                h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
267
            for(list=0; list<h->list_count; list++){
268
                if(USES_LIST(mb_type,list)){
269
                    uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
270
                    uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
271
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
272
                    for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
273
                        dst[0] = src[0];
274
                        dst[1] = src[1];
275
                        dst[2] = src[2];
276
                        dst[3] = src[3];
277
                    }
278
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
279
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
280
                    ref += h->b8_stride;
281
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
282
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
283
                }else{
284
                    fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
285
                    fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
286
                }
287
            }
288
        }
289
    }else{
290
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
291
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
292
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
293
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
294
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
295
    }
296

    
297
    if(IS_INTRA(mb_type)){
298
        h->topleft_samples_available=
299
        h->top_samples_available=
300
        h->left_samples_available= 0xFFFF;
301
        h->topright_samples_available= 0xEEEA;
302

    
303
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
304
            h->topleft_samples_available= 0xB3FF;
305
            h->top_samples_available= 0x33FF;
306
            h->topright_samples_available= 0x26EA;
307
        }
308
        for(i=0; i<2; i++){
309
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
310
                h->topleft_samples_available&= 0xDF5F;
311
                h->left_samples_available&= 0x5F5F;
312
            }
313
        }
314

    
315
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
316
            h->topleft_samples_available&= 0x7FFF;
317

    
318
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
319
            h->topright_samples_available&= 0xFBFF;
320

    
321
        if(IS_INTRA4x4(mb_type)){
322
            if(IS_INTRA4x4(top_type)){
323
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
324
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
325
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
326
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
327
            }else{
328
                int pred;
329
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
330
                    pred= -1;
331
                else{
332
                    pred= 2;
333
                }
334
                h->intra4x4_pred_mode_cache[4+8*0]=
335
                h->intra4x4_pred_mode_cache[5+8*0]=
336
                h->intra4x4_pred_mode_cache[6+8*0]=
337
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
338
            }
339
            for(i=0; i<2; i++){
340
                if(IS_INTRA4x4(left_type[i])){
341
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
342
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
343
                }else{
344
                    int pred;
345
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
346
                        pred= -1;
347
                    else{
348
                        pred= 2;
349
                    }
350
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
351
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
352
                }
353
            }
354
        }
355
    }
356

    
357

    
358
/*
359
0 . T T. T T T T
360
1 L . .L . . . .
361
2 L . .L . . . .
362
3 . T TL . . . .
363
4 L . .L . . . .
364
5 L . .. . . . .
365
*/
366
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
367
    if(top_type){
368
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
369
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
370
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
371
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
372

    
373
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
374
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
375

    
376
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
377
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
378

    
379
    }else{
380
        h->non_zero_count_cache[4+8*0]=
381
        h->non_zero_count_cache[5+8*0]=
382
        h->non_zero_count_cache[6+8*0]=
383
        h->non_zero_count_cache[7+8*0]=
384

    
385
        h->non_zero_count_cache[1+8*0]=
386
        h->non_zero_count_cache[2+8*0]=
387

    
388
        h->non_zero_count_cache[1+8*3]=
389
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
390

    
391
    }
392

    
393
    for (i=0; i<2; i++) {
394
        if(left_type[i]){
395
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
396
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
397
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
398
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
399
        }else{
400
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
401
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
402
            h->non_zero_count_cache[0+8*1 +   8*i]=
403
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
404
        }
405
    }
406

    
407
    if( h->pps.cabac ) {
408
        // top_cbp
409
        if(top_type) {
410
            h->top_cbp = h->cbp_table[top_xy];
411
        } else if(IS_INTRA(mb_type)) {
412
            h->top_cbp = 0x1C0;
413
        } else {
414
            h->top_cbp = 0;
415
        }
416
        // left_cbp
417
        if (left_type[0]) {
418
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
419
        } else if(IS_INTRA(mb_type)) {
420
            h->left_cbp = 0x1C0;
421
        } else {
422
            h->left_cbp = 0;
423
        }
424
        if (left_type[0]) {
425
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
426
        }
427
        if (left_type[1]) {
428
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
429
        }
430
    }
431

    
432
#if 1
433
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
434
        int list;
435
        for(list=0; list<h->list_count; list++){
436
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
437
                /*if(!h->mv_cache_clean[list]){
438
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
439
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
440
                    h->mv_cache_clean[list]= 1;
441
                }*/
442
                continue;
443
            }
444
            h->mv_cache_clean[list]= 0;
445

    
446
            if(USES_LIST(top_type, list)){
447
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
448
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
449
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
450
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
451
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
452
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
453
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
454
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
455
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
456
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
457
            }else{
458
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
459
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
460
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
461
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
462
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
463
            }
464

    
465
            for(i=0; i<2; i++){
466
                int cache_idx = scan8[0] - 1 + i*2*8;
467
                if(USES_LIST(left_type[i], list)){
468
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
469
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
470
                    *(uint32_t*)h->mv_cache[list][cache_idx  ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
471
                    *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
472
                    h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
473
                    h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
474
                }else{
475
                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
476
                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
477
                    h->ref_cache[list][cache_idx  ]=
478
                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
479
                }
480
            }
481

    
482
            if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
483
                continue;
484

    
485
            if(USES_LIST(topleft_type, list)){
486
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
487
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
488
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
489
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
490
            }else{
491
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
492
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
493
            }
494

    
495
            if(USES_LIST(topright_type, list)){
496
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
497
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
498
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
499
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
500
            }else{
501
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
502
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
503
            }
504

    
505
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
506
                continue;
507

    
508
            h->ref_cache[list][scan8[5 ]+1] =
509
            h->ref_cache[list][scan8[7 ]+1] =
510
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
511
            h->ref_cache[list][scan8[4 ]] =
512
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
513
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
514
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
515
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
516
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
517
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
518

    
519
            if( h->pps.cabac ) {
520
                /* XXX beurk, Load mvd */
521
                if(USES_LIST(top_type, list)){
522
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
523
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
524
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
525
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
526
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
527
                }else{
528
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
529
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
530
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
531
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
532
                }
533
                if(USES_LIST(left_type[0], list)){
534
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
535
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
536
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
537
                }else{
538
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
539
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
540
                }
541
                if(USES_LIST(left_type[1], list)){
542
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
543
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
544
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
545
                }else{
546
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
547
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
548
                }
549
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
550
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
551
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
552
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
553
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
554

    
555
                if(h->slice_type == B_TYPE){
556
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
557

    
558
                    if(IS_DIRECT(top_type)){
559
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
560
                    }else if(IS_8X8(top_type)){
561
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
562
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
563
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
564
                    }else{
565
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
566
                    }
567

    
568
                    if(IS_DIRECT(left_type[0]))
569
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
570
                    else if(IS_8X8(left_type[0]))
571
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
572
                    else
573
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
574

    
575
                    if(IS_DIRECT(left_type[1]))
576
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
577
                    else if(IS_8X8(left_type[1]))
578
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
579
                    else
580
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
581
                }
582
            }
583

    
584
            if(FRAME_MBAFF){
585
#define MAP_MVS\
586
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
587
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
588
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
589
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
590
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
591
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
592
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
593
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
594
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
595
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
596
                if(MB_FIELD){
597
#define MAP_F2F(idx, mb_type)\
598
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
599
                        h->ref_cache[list][idx] <<= 1;\
600
                        h->mv_cache[list][idx][1] /= 2;\
601
                        h->mvd_cache[list][idx][1] /= 2;\
602
                    }
603
                    MAP_MVS
604
#undef MAP_F2F
605
                }else{
606
#define MAP_F2F(idx, mb_type)\
607
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
608
                        h->ref_cache[list][idx] >>= 1;\
609
                        h->mv_cache[list][idx][1] <<= 1;\
610
                        h->mvd_cache[list][idx][1] <<= 1;\
611
                    }
612
                    MAP_MVS
613
#undef MAP_F2F
614
                }
615
            }
616
        }
617
    }
618
#endif
619

    
620
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
621
}
622

    
623
static inline void write_back_intra_pred_mode(H264Context *h){
624
    MpegEncContext * const s = &h->s;
625
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
626

    
627
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
628
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
629
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
630
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
631
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
632
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
633
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
634
}
635

    
636
/**
637
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
638
 */
639
static inline int check_intra4x4_pred_mode(H264Context *h){
640
    MpegEncContext * const s = &h->s;
641
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
642
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
643
    int i;
644

    
645
    if(!(h->top_samples_available&0x8000)){
646
        for(i=0; i<4; i++){
647
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
648
            if(status<0){
649
                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);
650
                return -1;
651
            } else if(status){
652
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
653
            }
654
        }
655
    }
656

    
657
    if(!(h->left_samples_available&0x8000)){
658
        for(i=0; i<4; i++){
659
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
660
            if(status<0){
661
                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);
662
                return -1;
663
            } else if(status){
664
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
665
            }
666
        }
667
    }
668

    
669
    return 0;
670
} //FIXME cleanup like next
671

    
672
/**
673
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
674
 */
675
static inline int check_intra_pred_mode(H264Context *h, int mode){
676
    MpegEncContext * const s = &h->s;
677
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
678
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
679

    
680
    if(mode > 6U) {
681
        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);
682
        return -1;
683
    }
684

    
685
    if(!(h->top_samples_available&0x8000)){
686
        mode= top[ mode ];
687
        if(mode<0){
688
            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);
689
            return -1;
690
        }
691
    }
692

    
693
    if(!(h->left_samples_available&0x8000)){
694
        mode= left[ mode ];
695
        if(mode<0){
696
            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);
697
            return -1;
698
        }
699
    }
700

    
701
    return mode;
702
}
703

    
704
/**
705
 * gets the predicted intra4x4 prediction mode.
706
 */
707
static inline int pred_intra_mode(H264Context *h, int n){
708
    const int index8= scan8[n];
709
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
710
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
711
    const int min= FFMIN(left, top);
712

    
713
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
714

    
715
    if(min<0) return DC_PRED;
716
    else      return min;
717
}
718

    
719
static inline void write_back_non_zero_count(H264Context *h){
720
    MpegEncContext * const s = &h->s;
721
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
722

    
723
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
724
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
725
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
726
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
727
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
728
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
729
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
730

    
731
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
732
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
733
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
734

    
735
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
736
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
737
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
738

    
739
    if(FRAME_MBAFF){
740
        // store all luma nnzs, for deblocking
741
        int v = 0, i;
742
        for(i=0; i<16; i++)
743
            v += (!!h->non_zero_count_cache[scan8[i]]) << i;
744
        *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
745
    }
746
}
747

    
748
/**
749
 * gets the predicted number of non zero coefficients.
750
 * @param n block index
751
 */
752
static inline int pred_non_zero_count(H264Context *h, int n){
753
    const int index8= scan8[n];
754
    const int left= h->non_zero_count_cache[index8 - 1];
755
    const int top = h->non_zero_count_cache[index8 - 8];
756
    int i= left + top;
757

    
758
    if(i<64) i= (i+1)>>1;
759

    
760
    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
761

    
762
    return i&31;
763
}
764

    
765
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
766
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
767
    MpegEncContext *s = &h->s;
768

    
769
    /* there is no consistent mapping of mvs to neighboring locations that will
770
     * make mbaff happy, so we can't move all this logic to fill_caches */
771
    if(FRAME_MBAFF){
772
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
773
        const int16_t *mv;
774
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
775
        *C = h->mv_cache[list][scan8[0]-2];
776

    
777
        if(!MB_FIELD
778
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
779
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
780
            if(IS_INTERLACED(mb_types[topright_xy])){
781
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
782
                const int x4 = X4, y4 = Y4;\
783
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
784
                if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\
785
                    return LIST_NOT_USED;\
786
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
787
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
788
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
789
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
790

    
791
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
792
            }
793
        }
794
        if(topright_ref == PART_NOT_AVAILABLE
795
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
796
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
797
            if(!MB_FIELD
798
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
799
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
800
            }
801
            if(MB_FIELD
802
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
803
               && i >= scan8[0]+8){
804
                // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
805
                SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
806
            }
807
        }
808
#undef SET_DIAG_MV
809
    }
810

    
811
    if(topright_ref != PART_NOT_AVAILABLE){
812
        *C= h->mv_cache[list][ i - 8 + part_width ];
813
        return topright_ref;
814
    }else{
815
        tprintf(s->avctx, "topright MV not available\n");
816

    
817
        *C= h->mv_cache[list][ i - 8 - 1 ];
818
        return h->ref_cache[list][ i - 8 - 1 ];
819
    }
820
}
821

    
822
/**
823
 * gets the predicted MV.
824
 * @param n the block index
825
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
826
 * @param mx the x component of the predicted motion vector
827
 * @param my the y component of the predicted motion vector
828
 */
829
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
830
    const int index8= scan8[n];
831
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
832
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
833
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
834
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
835
    const int16_t * C;
836
    int diagonal_ref, match_count;
837

    
838
    assert(part_width==1 || part_width==2 || part_width==4);
839

    
840
/* mv_cache
841
  B . . A T T T T
842
  U . . L . . , .
843
  U . . L . . . .
844
  U . . L . . , .
845
  . . . L . . . .
846
*/
847

    
848
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
849
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
850
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
851
    if(match_count > 1){ //most common
852
        *mx= mid_pred(A[0], B[0], C[0]);
853
        *my= mid_pred(A[1], B[1], C[1]);
854
    }else if(match_count==1){
855
        if(left_ref==ref){
856
            *mx= A[0];
857
            *my= A[1];
858
        }else if(top_ref==ref){
859
            *mx= B[0];
860
            *my= B[1];
861
        }else{
862
            *mx= C[0];
863
            *my= C[1];
864
        }
865
    }else{
866
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
867
            *mx= A[0];
868
            *my= A[1];
869
        }else{
870
            *mx= mid_pred(A[0], B[0], C[0]);
871
            *my= mid_pred(A[1], B[1], C[1]);
872
        }
873
    }
874

    
875
    tprintf(h->s.avctx, "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1],                    diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
876
}
877

    
878
/**
879
 * gets the directionally predicted 16x8 MV.
880
 * @param n the block index
881
 * @param mx the x component of the predicted motion vector
882
 * @param my the y component of the predicted motion vector
883
 */
884
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
885
    if(n==0){
886
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
887
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
888

    
889
        tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
890

    
891
        if(top_ref == ref){
892
            *mx= B[0];
893
            *my= B[1];
894
            return;
895
        }
896
    }else{
897
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
898
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
899

    
900
        tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
901

    
902
        if(left_ref == ref){
903
            *mx= A[0];
904
            *my= A[1];
905
            return;
906
        }
907
    }
908

    
909
    //RARE
910
    pred_motion(h, n, 4, list, ref, mx, my);
911
}
912

    
913
/**
914
 * gets the directionally predicted 8x16 MV.
915
 * @param n the block index
916
 * @param mx the x component of the predicted motion vector
917
 * @param my the y component of the predicted motion vector
918
 */
919
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
920
    if(n==0){
921
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
922
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
923

    
924
        tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
925

    
926
        if(left_ref == ref){
927
            *mx= A[0];
928
            *my= A[1];
929
            return;
930
        }
931
    }else{
932
        const int16_t * C;
933
        int diagonal_ref;
934

    
935
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
936

    
937
        tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
938

    
939
        if(diagonal_ref == ref){
940
            *mx= C[0];
941
            *my= C[1];
942
            return;
943
        }
944
    }
945

    
946
    //RARE
947
    pred_motion(h, n, 2, list, ref, mx, my);
948
}
949

    
950
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
951
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
952
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
953

    
954
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
955

    
956
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
957
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
958
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
959

    
960
        *mx = *my = 0;
961
        return;
962
    }
963

    
964
    pred_motion(h, 0, 4, 0, 0, mx, my);
965

    
966
    return;
967
}
968

    
969
static inline void direct_dist_scale_factor(H264Context * const h){
970
    const int poc = h->s.current_picture_ptr->poc;
971
    const int poc1 = h->ref_list[1][0].poc;
972
    int i;
973
    for(i=0; i<h->ref_count[0]; i++){
974
        int poc0 = h->ref_list[0][i].poc;
975
        int td = av_clip(poc1 - poc0, -128, 127);
976
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
977
            h->dist_scale_factor[i] = 256;
978
        }else{
979
            int tb = av_clip(poc - poc0, -128, 127);
980
            int tx = (16384 + (FFABS(td) >> 1)) / td;
981
            h->dist_scale_factor[i] = av_clip((tb*tx + 32) >> 6, -1024, 1023);
982
        }
983
    }
984
    if(FRAME_MBAFF){
985
        for(i=0; i<h->ref_count[0]; i++){
986
            h->dist_scale_factor_field[2*i] =
987
            h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
988
        }
989
    }
990
}
991
static inline void direct_ref_list_init(H264Context * const h){
992
    MpegEncContext * const s = &h->s;
993
    Picture * const ref1 = &h->ref_list[1][0];
994
    Picture * const cur = s->current_picture_ptr;
995
    int list, i, j;
996
    if(cur->pict_type == I_TYPE)
997
        cur->ref_count[0] = 0;
998
    if(cur->pict_type != B_TYPE)
999
        cur->ref_count[1] = 0;
1000
    for(list=0; list<2; list++){
1001
        cur->ref_count[list] = h->ref_count[list];
1002
        for(j=0; j<h->ref_count[list]; j++)
1003
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1004
    }
1005
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1006
        return;
1007
    for(list=0; list<2; list++){
1008
        for(i=0; i<ref1->ref_count[list]; i++){
1009
            const int poc = ref1->ref_poc[list][i];
1010
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
1011
            for(j=0; j<h->ref_count[list]; j++)
1012
                if(h->ref_list[list][j].poc == poc){
1013
                    h->map_col_to_list0[list][i] = j;
1014
                    break;
1015
                }
1016
        }
1017
    }
1018
    if(FRAME_MBAFF){
1019
        for(list=0; list<2; list++){
1020
            for(i=0; i<ref1->ref_count[list]; i++){
1021
                j = h->map_col_to_list0[list][i];
1022
                h->map_col_to_list0_field[list][2*i] = 2*j;
1023
                h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
1024
            }
1025
        }
1026
    }
1027
}
1028

    
1029
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1030
    MpegEncContext * const s = &h->s;
1031
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1032
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1033
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1034
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1035
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1036
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1037
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1038
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1039
    const int is_b8x8 = IS_8X8(*mb_type);
1040
    unsigned int sub_mb_type;
1041
    int i8, i4;
1042

    
1043
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
1044
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1045
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1046
         * so we know exactly what block size to use */
1047
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1048
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1049
    }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
1050
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1051
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1052
    }else{
1053
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1054
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1055
    }
1056
    if(!is_b8x8)
1057
        *mb_type |= MB_TYPE_DIRECT2;
1058
    if(MB_FIELD)
1059
        *mb_type |= MB_TYPE_INTERLACED;
1060

    
1061
    tprintf(s->avctx, "mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
1062

    
1063
    if(h->direct_spatial_mv_pred){
1064
        int ref[2];
1065
        int mv[2][2];
1066
        int list;
1067

    
1068
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1069

    
1070
        /* ref = min(neighbors) */
1071
        for(list=0; list<2; list++){
1072
            int refa = h->ref_cache[list][scan8[0] - 1];
1073
            int refb = h->ref_cache[list][scan8[0] - 8];
1074
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1075
            if(refc == -2)
1076
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1077
            ref[list] = refa;
1078
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1079
                ref[list] = refb;
1080
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1081
                ref[list] = refc;
1082
            if(ref[list] < 0)
1083
                ref[list] = -1;
1084
        }
1085

    
1086
        if(ref[0] < 0 && ref[1] < 0){
1087
            ref[0] = ref[1] = 0;
1088
            mv[0][0] = mv[0][1] =
1089
            mv[1][0] = mv[1][1] = 0;
1090
        }else{
1091
            for(list=0; list<2; list++){
1092
                if(ref[list] >= 0)
1093
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1094
                else
1095
                    mv[list][0] = mv[list][1] = 0;
1096
            }
1097
        }
1098

    
1099
        if(ref[1] < 0){
1100
            *mb_type &= ~MB_TYPE_P0L1;
1101
            sub_mb_type &= ~MB_TYPE_P0L1;
1102
        }else if(ref[0] < 0){
1103
            *mb_type &= ~MB_TYPE_P0L0;
1104
            sub_mb_type &= ~MB_TYPE_P0L0;
1105
        }
1106

    
1107
        if(IS_16X16(*mb_type)){
1108
            int a=0, b=0;
1109

    
1110
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1111
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1112
            if(!IS_INTRA(mb_type_col)
1113
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1114
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1115
                       && (h->x264_build>33 || !h->x264_build)))){
1116
                if(ref[0] > 0)
1117
                    a= pack16to32(mv[0][0],mv[0][1]);
1118
                if(ref[1] > 0)
1119
                    b= pack16to32(mv[1][0],mv[1][1]);
1120
            }else{
1121
                a= pack16to32(mv[0][0],mv[0][1]);
1122
                b= pack16to32(mv[1][0],mv[1][1]);
1123
            }
1124
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1125
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1126
        }else{
1127
            for(i8=0; i8<4; i8++){
1128
                const int x8 = i8&1;
1129
                const int y8 = i8>>1;
1130

    
1131
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1132
                    continue;
1133
                h->sub_mb_type[i8] = sub_mb_type;
1134

    
1135
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1136
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1137
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1138
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1139

    
1140
                /* col_zero_flag */
1141
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0
1142
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1143
                                                  && (h->x264_build>33 || !h->x264_build)))){
1144
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1145
                    if(IS_SUB_8X8(sub_mb_type)){
1146
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1147
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1148
                            if(ref[0] == 0)
1149
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1150
                            if(ref[1] == 0)
1151
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1152
                        }
1153
                    }else
1154
                    for(i4=0; i4<4; i4++){
1155
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1156
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1157
                            if(ref[0] == 0)
1158
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1159
                            if(ref[1] == 0)
1160
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1161
                        }
1162
                    }
1163
                }
1164
            }
1165
        }
1166
    }else{ /* direct temporal mv pred */
1167
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1168
        const int *dist_scale_factor = h->dist_scale_factor;
1169

    
1170
        if(FRAME_MBAFF){
1171
            if(IS_INTERLACED(*mb_type)){
1172
                map_col_to_list0[0] = h->map_col_to_list0_field[0];
1173
                map_col_to_list0[1] = h->map_col_to_list0_field[1];
1174
                dist_scale_factor = h->dist_scale_factor_field;
1175
            }
1176
            if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1177
                /* FIXME assumes direct_8x8_inference == 1 */
1178
                const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1179
                int mb_types_col[2];
1180
                int y_shift;
1181

    
1182
                *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1183
                         | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1184
                         | (*mb_type & MB_TYPE_INTERLACED);
1185
                sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1186

    
1187
                if(IS_INTERLACED(*mb_type)){
1188
                    /* frame to field scaling */
1189
                    mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1190
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1191
                    if(s->mb_y&1){
1192
                        l1ref0 -= 2*h->b8_stride;
1193
                        l1ref1 -= 2*h->b8_stride;
1194
                        l1mv0 -= 4*h->b_stride;
1195
                        l1mv1 -= 4*h->b_stride;
1196
                    }
1197
                    y_shift = 0;
1198

    
1199
                    if(   (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1200
                       && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1201
                       && !is_b8x8)
1202
                        *mb_type |= MB_TYPE_16x8;
1203
                    else
1204
                        *mb_type |= MB_TYPE_8x8;
1205
                }else{
1206
                    /* field to frame scaling */
1207
                    /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
1208
                     * but in MBAFF, top and bottom POC are equal */
1209
                    int dy = (s->mb_y&1) ? 1 : 2;
1210
                    mb_types_col[0] =
1211
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1212
                    l1ref0 += dy*h->b8_stride;
1213
                    l1ref1 += dy*h->b8_stride;
1214
                    l1mv0 += 2*dy*h->b_stride;
1215
                    l1mv1 += 2*dy*h->b_stride;
1216
                    y_shift = 2;
1217

    
1218
                    if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1219
                       && !is_b8x8)
1220
                        *mb_type |= MB_TYPE_16x16;
1221
                    else
1222
                        *mb_type |= MB_TYPE_8x8;
1223
                }
1224

    
1225
                for(i8=0; i8<4; i8++){
1226
                    const int x8 = i8&1;
1227
                    const int y8 = i8>>1;
1228
                    int ref0, scale;
1229
                    const int16_t (*l1mv)[2]= l1mv0;
1230

    
1231
                    if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1232
                        continue;
1233
                    h->sub_mb_type[i8] = sub_mb_type;
1234

    
1235
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1236
                    if(IS_INTRA(mb_types_col[y8])){
1237
                        fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1238
                        fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1239
                        fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1240
                        continue;
1241
                    }
1242

    
1243
                    ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1244
                    if(ref0 >= 0)
1245
                        ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1246
                    else{
1247
                        ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1248
                        l1mv= l1mv1;
1249
                    }
1250
                    scale = dist_scale_factor[ref0];
1251
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1252

    
1253
                    {
1254
                        const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
1255
                        int my_col = (mv_col[1]<<y_shift)/2;
1256
                        int mx = (scale * mv_col[0] + 128) >> 8;
1257
                        int my = (scale * my_col + 128) >> 8;
1258
                        fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1259
                        fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1260
                    }
1261
                }
1262
                return;
1263
            }
1264
        }
1265

    
1266
        /* one-to-one mv scaling */
1267

    
1268
        if(IS_16X16(*mb_type)){
1269
            int ref, mv0, mv1;
1270

    
1271
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1272
            if(IS_INTRA(mb_type_col)){
1273
                ref=mv0=mv1=0;
1274
            }else{
1275
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1276
                                                : map_col_to_list0[1][l1ref1[0]];
1277
                const int scale = dist_scale_factor[ref0];
1278
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1279
                int mv_l0[2];
1280
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1281
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1282
                ref= ref0;
1283
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
1284
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1285
            }
1286
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1287
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1288
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1289
        }else{
1290
            for(i8=0; i8<4; i8++){
1291
                const int x8 = i8&1;
1292
                const int y8 = i8>>1;
1293
                int ref0, scale;
1294
                const int16_t (*l1mv)[2]= l1mv0;
1295

    
1296
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1297
                    continue;
1298
                h->sub_mb_type[i8] = sub_mb_type;
1299
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1300
                if(IS_INTRA(mb_type_col)){
1301
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1302
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1303
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1304
                    continue;
1305
                }
1306

    
1307
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1308
                if(ref0 >= 0)
1309
                    ref0 = map_col_to_list0[0][ref0];
1310
                else{
1311
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1312
                    l1mv= l1mv1;
1313
                }
1314
                scale = dist_scale_factor[ref0];
1315

    
1316
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1317
                if(IS_SUB_8X8(sub_mb_type)){
1318
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1319
                    int mx = (scale * mv_col[0] + 128) >> 8;
1320
                    int my = (scale * mv_col[1] + 128) >> 8;
1321
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1322
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1323
                }else
1324
                for(i4=0; i4<4; i4++){
1325
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1326
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1327
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1328
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1329
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1330
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1331
                }
1332
            }
1333
        }
1334
    }
1335
}
1336

    
1337
static inline void write_back_motion(H264Context *h, int mb_type){
1338
    MpegEncContext * const s = &h->s;
1339
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1340
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1341
    int list;
1342

    
1343
    if(!USES_LIST(mb_type, 0))
1344
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1345

    
1346
    for(list=0; list<h->list_count; list++){
1347
        int y;
1348
        if(!USES_LIST(mb_type, list))
1349
            continue;
1350

    
1351
        for(y=0; y<4; y++){
1352
            *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
1353
            *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
1354
        }
1355
        if( h->pps.cabac ) {
1356
            if(IS_SKIP(mb_type))
1357
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1358
            else
1359
            for(y=0; y<4; y++){
1360
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1361
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1362
            }
1363
        }
1364

    
1365
        {
1366
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1367
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1368
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1369
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1370
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1371
        }
1372
    }
1373

    
1374
    if(h->slice_type == B_TYPE && h->pps.cabac){
1375
        if(IS_8X8(mb_type)){
1376
            uint8_t *direct_table = &h->direct_table[b8_xy];
1377
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1378
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1379
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1380
        }
1381
    }
1382
}
1383

    
1384
/**
1385
 * Decodes a network abstraction layer unit.
1386
 * @param consumed is the number of bytes used as input
1387
 * @param length is the length of the array
1388
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1389
 * @returns decoded bytes, might be src+1 if no escapes
1390
 */
1391
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1392
    int i, si, di;
1393
    uint8_t *dst;
1394
    int bufidx;
1395

    
1396
//    src[0]&0x80;                //forbidden bit
1397
    h->nal_ref_idc= src[0]>>5;
1398
    h->nal_unit_type= src[0]&0x1F;
1399

    
1400
    src++; length--;
1401
#if 0
1402
    for(i=0; i<length; i++)
1403
        printf("%2X ", src[i]);
1404
#endif
1405
    for(i=0; i+1<length; i+=2){
1406
        if(src[i]) continue;
1407
        if(i>0 && src[i-1]==0) i--;
1408
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1409
            if(src[i+2]!=3){
1410
                /* startcode, so we must be past the end */
1411
                length=i;
1412
            }
1413
            break;
1414
        }
1415
    }
1416

    
1417
    if(i>=length-1){ //no escaped 0
1418
        *dst_length= length;
1419
        *consumed= length+1; //+1 for the header
1420
        return src;
1421
    }
1422

    
1423
    bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1424
    h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length);
1425
    dst= h->rbsp_buffer[bufidx];
1426

    
1427
    if (dst == NULL){
1428
        return NULL;
1429
    }
1430

    
1431
//printf("decoding esc\n");
1432
    si=di=0;
1433
    while(si<length){
1434
        //remove escapes (very rare 1:2^22)
1435
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1436
            if(src[si+2]==3){ //escape
1437
                dst[di++]= 0;
1438
                dst[di++]= 0;
1439
                si+=3;
1440
                continue;
1441
            }else //next start code
1442
                break;
1443
        }
1444

    
1445
        dst[di++]= src[si++];
1446
    }
1447

    
1448
    *dst_length= di;
1449
    *consumed= si + 1;//+1 for the header
1450
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1451
    return dst;
1452
}
1453

    
1454
/**
1455
 * identifies the exact end of the bitstream
1456
 * @return the length of the trailing, or 0 if damaged
1457
 */
1458
static int decode_rbsp_trailing(H264Context *h, uint8_t *src){
1459
    int v= *src;
1460
    int r;
1461

    
1462
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1463

    
1464
    for(r=1; r<9; r++){
1465
        if(v&1) return r;
1466
        v>>=1;
1467
    }
1468
    return 0;
1469
}
1470

    
1471
/**
1472
 * idct tranforms the 16 dc values and dequantize them.
1473
 * @param qp quantization parameter
1474
 */
1475
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1476
#define stride 16
1477
    int i;
1478
    int temp[16]; //FIXME check if this is a good idea
1479
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1480
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1481

    
1482
//memset(block, 64, 2*256);
1483
//return;
1484
    for(i=0; i<4; i++){
1485
        const int offset= y_offset[i];
1486
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1487
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1488
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1489
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1490

    
1491
        temp[4*i+0]= z0+z3;
1492
        temp[4*i+1]= z1+z2;
1493
        temp[4*i+2]= z1-z2;
1494
        temp[4*i+3]= z0-z3;
1495
    }
1496

    
1497
    for(i=0; i<4; i++){
1498
        const int offset= x_offset[i];
1499
        const int z0= temp[4*0+i] + temp[4*2+i];
1500
        const int z1= temp[4*0+i] - temp[4*2+i];
1501
        const int z2= temp[4*1+i] - temp[4*3+i];
1502
        const int z3= temp[4*1+i] + temp[4*3+i];
1503

    
1504
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1505
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1506
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1507
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1508
    }
1509
}
1510

    
1511
#if 0
1512
/**
1513
 * dct tranforms the 16 dc values.
1514
 * @param qp quantization parameter ??? FIXME
1515
 */
1516
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1517
//    const int qmul= dequant_coeff[qp][0];
1518
    int i;
1519
    int temp[16]; //FIXME check if this is a good idea
1520
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1521
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1522

1523
    for(i=0; i<4; i++){
1524
        const int offset= y_offset[i];
1525
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1526
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1527
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1528
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1529

1530
        temp[4*i+0]= z0+z3;
1531
        temp[4*i+1]= z1+z2;
1532
        temp[4*i+2]= z1-z2;
1533
        temp[4*i+3]= z0-z3;
1534
    }
1535

1536
    for(i=0; i<4; i++){
1537
        const int offset= x_offset[i];
1538
        const int z0= temp[4*0+i] + temp[4*2+i];
1539
        const int z1= temp[4*0+i] - temp[4*2+i];
1540
        const int z2= temp[4*1+i] - temp[4*3+i];
1541
        const int z3= temp[4*1+i] + temp[4*3+i];
1542

1543
        block[stride*0 +offset]= (z0 + z3)>>1;
1544
        block[stride*2 +offset]= (z1 + z2)>>1;
1545
        block[stride*8 +offset]= (z1 - z2)>>1;
1546
        block[stride*10+offset]= (z0 - z3)>>1;
1547
    }
1548
}
1549
#endif
1550

    
1551
#undef xStride
1552
#undef stride
1553

    
1554
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1555
    const int stride= 16*2;
1556
    const int xStride= 16;
1557
    int a,b,c,d,e;
1558

    
1559
    a= block[stride*0 + xStride*0];
1560
    b= block[stride*0 + xStride*1];
1561
    c= block[stride*1 + xStride*0];
1562
    d= block[stride*1 + xStride*1];
1563

    
1564
    e= a-b;
1565
    a= a+b;
1566
    b= c-d;
1567
    c= c+d;
1568

    
1569
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1570
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1571
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1572
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1573
}
1574

    
1575
#if 0
1576
static void chroma_dc_dct_c(DCTELEM *block){
1577
    const int stride= 16*2;
1578
    const int xStride= 16;
1579
    int a,b,c,d,e;
1580

1581
    a= block[stride*0 + xStride*0];
1582
    b= block[stride*0 + xStride*1];
1583
    c= block[stride*1 + xStride*0];
1584
    d= block[stride*1 + xStride*1];
1585

1586
    e= a-b;
1587
    a= a+b;
1588
    b= c-d;
1589
    c= c+d;
1590

1591
    block[stride*0 + xStride*0]= (a+c);
1592
    block[stride*0 + xStride*1]= (e+b);
1593
    block[stride*1 + xStride*0]= (a-c);
1594
    block[stride*1 + xStride*1]= (e-b);
1595
}
1596
#endif
1597

    
1598
/**
1599
 * gets the chroma qp.
1600
 */
1601
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1602
    return h->pps.chroma_qp_table[t][qscale & 0xff];
1603
}
1604

    
1605
//FIXME need to check that this does not overflow signed 32 bit for low qp, i am not sure, it's very close
1606
//FIXME check that gcc inlines this (and optimizes intra & separate_dc stuff away)
1607
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int separate_dc){
1608
    int i;
1609
    const int * const quant_table= quant_coeff[qscale];
1610
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1611
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1612
    const unsigned int threshold2= (threshold1<<1);
1613
    int last_non_zero;
1614

    
1615
    if(separate_dc){
1616
        if(qscale<=18){
1617
            //avoid overflows
1618
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1619
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1620
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1621

    
1622
            int level= block[0]*quant_coeff[qscale+18][0];
1623
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1624
                if(level>0){
1625
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1626
                    block[0]= level;
1627
                }else{
1628
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1629
                    block[0]= -level;
1630
                }
1631
//                last_non_zero = i;
1632
            }else{
1633
                block[0]=0;
1634
            }
1635
        }else{
1636
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1637
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1638
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1639

    
1640
            int level= block[0]*quant_table[0];
1641
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1642
                if(level>0){
1643
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1644
                    block[0]= level;
1645
                }else{
1646
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1647
                    block[0]= -level;
1648
                }
1649
//                last_non_zero = i;
1650
            }else{
1651
                block[0]=0;
1652
            }
1653
        }
1654
        last_non_zero= 0;
1655
        i=1;
1656
    }else{
1657
        last_non_zero= -1;
1658
        i=0;
1659
    }
1660

    
1661
    for(; i<16; i++){
1662
        const int j= scantable[i];
1663
        int level= block[j]*quant_table[j];
1664

    
1665
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1666
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1667
        if(((unsigned)(level+threshold1))>threshold2){
1668
            if(level>0){
1669
                level= (bias + level)>>QUANT_SHIFT;
1670
                block[j]= level;
1671
            }else{
1672
                level= (bias - level)>>QUANT_SHIFT;
1673
                block[j]= -level;
1674
            }
1675
            last_non_zero = i;
1676
        }else{
1677
            block[j]=0;
1678
        }
1679
    }
1680

    
1681
    return last_non_zero;
1682
}
1683

    
1684
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1685
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1686
                           int src_x_offset, int src_y_offset,
1687
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1688
    MpegEncContext * const s = &h->s;
1689
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1690
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1691
    const int luma_xy= (mx&3) + ((my&3)<<2);
1692
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1693
    uint8_t * src_cb, * src_cr;
1694
    int extra_width= h->emu_edge_width;
1695
    int extra_height= h->emu_edge_height;
1696
    int emu=0;
1697
    const int full_mx= mx>>2;
1698
    const int full_my= my>>2;
1699
    const int pic_width  = 16*s->mb_width;
1700
    const int pic_height = 16*s->mb_height >> MB_MBAFF;
1701

    
1702
    if(!pic->data[0]) //FIXME this is unacceptable, some senseable error concealment must be done for missing reference frames
1703
        return;
1704

    
1705
    if(mx&7) extra_width -= 3;
1706
    if(my&7) extra_height -= 3;
1707

    
1708
    if(   full_mx < 0-extra_width
1709
       || full_my < 0-extra_height
1710
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
1711
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
1712
        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);
1713
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1714
        emu=1;
1715
    }
1716

    
1717
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1718
    if(!square){
1719
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1720
    }
1721

    
1722
    if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1723

    
1724
    if(MB_MBAFF){
1725
        // chroma offset when predicting from a field of opposite parity
1726
        my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));
1727
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1728
    }
1729
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1730
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1731

    
1732
    if(emu){
1733
        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);
1734
            src_cb= s->edge_emu_buffer;
1735
    }
1736
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1737

    
1738
    if(emu){
1739
        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);
1740
            src_cr= s->edge_emu_buffer;
1741
    }
1742
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1743
}
1744

    
1745
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1746
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1747
                           int x_offset, int y_offset,
1748
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1749
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1750
                           int list0, int list1){
1751
    MpegEncContext * const s = &h->s;
1752
    qpel_mc_func *qpix_op=  qpix_put;
1753
    h264_chroma_mc_func chroma_op= chroma_put;
1754

    
1755
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1756
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1757
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1758
    x_offset += 8*s->mb_x;
1759
    y_offset += 8*(s->mb_y >> MB_MBAFF);
1760

    
1761
    if(list0){
1762
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1763
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1764
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1765
                           qpix_op, chroma_op);
1766

    
1767
        qpix_op=  qpix_avg;
1768
        chroma_op= chroma_avg;
1769
    }
1770

    
1771
    if(list1){
1772
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1773
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1774
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1775
                           qpix_op, chroma_op);
1776
    }
1777
}
1778

    
1779
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1780
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1781
                           int x_offset, int y_offset,
1782
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1783
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1784
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1785
                           int list0, int list1){
1786
    MpegEncContext * const s = &h->s;
1787

    
1788
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1789
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1790
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1791
    x_offset += 8*s->mb_x;
1792
    y_offset += 8*(s->mb_y >> MB_MBAFF);
1793

    
1794
    if(list0 && list1){
1795
        /* don't optimize for luma-only case, since B-frames usually
1796
         * use implicit weights => chroma too. */
1797
        uint8_t *tmp_cb = s->obmc_scratchpad;
1798
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1799
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1800
        int refn0 = h->ref_cache[0][ scan8[n] ];
1801
        int refn1 = h->ref_cache[1][ scan8[n] ];
1802

    
1803
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1804
                    dest_y, dest_cb, dest_cr,
1805
                    x_offset, y_offset, qpix_put, chroma_put);
1806
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1807
                    tmp_y, tmp_cb, tmp_cr,
1808
                    x_offset, y_offset, qpix_put, chroma_put);
1809

    
1810
        if(h->use_weight == 2){
1811
            int weight0 = h->implicit_weight[refn0][refn1];
1812
            int weight1 = 64 - weight0;
1813
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
1814
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1815
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1816
        }else{
1817
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1818
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1819
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1820
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1821
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1822
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1823
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1824
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1825
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1826
        }
1827
    }else{
1828
        int list = list1 ? 1 : 0;
1829
        int refn = h->ref_cache[list][ scan8[n] ];
1830
        Picture *ref= &h->ref_list[list][refn];
1831
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1832
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
1833
                    qpix_put, chroma_put);
1834

    
1835
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1836
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
1837
        if(h->use_weight_chroma){
1838
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1839
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1840
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1841
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1842
        }
1843
    }
1844
}
1845

    
1846
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1847
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1848
                           int x_offset, int y_offset,
1849
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1850
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1851
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1852
                           int list0, int list1){
1853
    if((h->use_weight==2 && list0 && list1
1854
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1855
       || h->use_weight==1)
1856
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1857
                         x_offset, y_offset, qpix_put, chroma_put,
1858
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1859
    else
1860
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1861
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1862
}
1863

    
1864
static inline void prefetch_motion(H264Context *h, int list){
1865
    /* fetch pixels for estimated mv 4 macroblocks ahead
1866
     * optimized for 64byte cache lines */
1867
    MpegEncContext * const s = &h->s;
1868
    const int refn = h->ref_cache[list][scan8[0]];
1869
    if(refn >= 0){
1870
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1871
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1872
        uint8_t **src= h->ref_list[list][refn].data;
1873
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1874
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
1875
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1876
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1877
    }
1878
}
1879

    
1880
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1881
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1882
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1883
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1884
    MpegEncContext * const s = &h->s;
1885
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1886
    const int mb_type= s->current_picture.mb_type[mb_xy];
1887

    
1888
    assert(IS_INTER(mb_type));
1889

    
1890
    prefetch_motion(h, 0);
1891

    
1892
    if(IS_16X16(mb_type)){
1893
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1894
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1895
                &weight_op[0], &weight_avg[0],
1896
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1897
    }else if(IS_16X8(mb_type)){
1898
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1899
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1900
                &weight_op[1], &weight_avg[1],
1901
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1902
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1903
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1904
                &weight_op[1], &weight_avg[1],
1905
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1906
    }else if(IS_8X16(mb_type)){
1907
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1908
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1909
                &weight_op[2], &weight_avg[2],
1910
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1911
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1912
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1913
                &weight_op[2], &weight_avg[2],
1914
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1915
    }else{
1916
        int i;
1917

    
1918
        assert(IS_8X8(mb_type));
1919

    
1920
        for(i=0; i<4; i++){
1921
            const int sub_mb_type= h->sub_mb_type[i];
1922
            const int n= 4*i;
1923
            int x_offset= (i&1)<<2;
1924
            int y_offset= (i&2)<<1;
1925

    
1926
            if(IS_SUB_8X8(sub_mb_type)){
1927
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1928
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1929
                    &weight_op[3], &weight_avg[3],
1930
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1931
            }else if(IS_SUB_8X4(sub_mb_type)){
1932
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1933
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1934
                    &weight_op[4], &weight_avg[4],
1935
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1936
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1937
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1938
                    &weight_op[4], &weight_avg[4],
1939
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1940
            }else if(IS_SUB_4X8(sub_mb_type)){
1941
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1942
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1943
                    &weight_op[5], &weight_avg[5],
1944
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1945
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1946
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1947
                    &weight_op[5], &weight_avg[5],
1948
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1949
            }else{
1950
                int j;
1951
                assert(IS_SUB_4X4(sub_mb_type));
1952
                for(j=0; j<4; j++){
1953
                    int sub_x_offset= x_offset + 2*(j&1);
1954
                    int sub_y_offset= y_offset +   (j&2);
1955
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1956
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1957
                        &weight_op[6], &weight_avg[6],
1958
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1959
                }
1960
            }
1961
        }
1962
    }
1963

    
1964
    prefetch_motion(h, 1);
1965
}
1966

    
1967
static void decode_init_vlc(void){
1968
    static int done = 0;
1969

    
1970
    if (!done) {
1971
        int i;
1972
        done = 1;
1973

    
1974
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1975
                 &chroma_dc_coeff_token_len [0], 1, 1,
1976
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
1977

    
1978
        for(i=0; i<4; i++){
1979
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1980
                     &coeff_token_len [i][0], 1, 1,
1981
                     &coeff_token_bits[i][0], 1, 1, 1);
1982
        }
1983

    
1984
        for(i=0; i<3; i++){
1985
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1986
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
1987
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
1988
        }
1989
        for(i=0; i<15; i++){
1990
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
1991
                     &total_zeros_len [i][0], 1, 1,
1992
                     &total_zeros_bits[i][0], 1, 1, 1);
1993
        }
1994

    
1995
        for(i=0; i<6; i++){
1996
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
1997
                     &run_len [i][0], 1, 1,
1998
                     &run_bits[i][0], 1, 1, 1);
1999
        }
2000
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2001
                 &run_len [6][0], 1, 1,
2002
                 &run_bits[6][0], 1, 1, 1);
2003
    }
2004
}
2005

    
2006
static void free_tables(H264Context *h){
2007
    int i;
2008
    H264Context *hx;
2009
    av_freep(&h->intra4x4_pred_mode);
2010
    av_freep(&h->chroma_pred_mode_table);
2011
    av_freep(&h->cbp_table);
2012
    av_freep(&h->mvd_table[0]);
2013
    av_freep(&h->mvd_table[1]);
2014
    av_freep(&h->direct_table);
2015
    av_freep(&h->non_zero_count);
2016
    av_freep(&h->slice_table_base);
2017
    h->slice_table= NULL;
2018

    
2019
    av_freep(&h->mb2b_xy);
2020
    av_freep(&h->mb2b8_xy);
2021

    
2022
    for(i = 0; i < MAX_SPS_COUNT; i++)
2023
        av_freep(h->sps_buffers + i);
2024

    
2025
    for(i = 0; i < MAX_PPS_COUNT; i++)
2026
        av_freep(h->pps_buffers + i);
2027

    
2028
    for(i = 0; i < h->s.avctx->thread_count; i++) {
2029
        hx = h->thread_context[i];
2030
        if(!hx) continue;
2031
        av_freep(&hx->top_borders[1]);
2032
        av_freep(&hx->top_borders[0]);
2033
        av_freep(&hx->s.obmc_scratchpad);
2034
        av_freep(&hx->s.allocated_edge_emu_buffer);
2035
    }
2036
}
2037

    
2038
static void init_dequant8_coeff_table(H264Context *h){
2039
    int i,q,x;
2040
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2041
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
2042
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
2043

    
2044
    for(i=0; i<2; i++ ){
2045
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2046
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
2047
            break;
2048
        }
2049

    
2050
        for(q=0; q<52; q++){
2051
            int shift = ff_div6[q];
2052
            int idx = ff_rem6[q];
2053
            for(x=0; x<64; x++)
2054
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2055
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2056
                    h->pps.scaling_matrix8[i][x]) << shift;
2057
        }
2058
    }
2059
}
2060

    
2061
static void init_dequant4_coeff_table(H264Context *h){
2062
    int i,j,q,x;
2063
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2064
    for(i=0; i<6; i++ ){
2065
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
2066
        for(j=0; j<i; j++){
2067
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2068
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2069
                break;
2070
            }
2071
        }
2072
        if(j<i)
2073
            continue;
2074

    
2075
        for(q=0; q<52; q++){
2076
            int shift = ff_div6[q] + 2;
2077
            int idx = ff_rem6[q];
2078
            for(x=0; x<16; x++)
2079
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2080
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2081
                    h->pps.scaling_matrix4[i][x]) << shift;
2082
        }
2083
    }
2084
}
2085

    
2086
static void init_dequant_tables(H264Context *h){
2087
    int i,x;
2088
    init_dequant4_coeff_table(h);
2089
    if(h->pps.transform_8x8_mode)
2090
        init_dequant8_coeff_table(h);
2091
    if(h->sps.transform_bypass){
2092
        for(i=0; i<6; i++)
2093
            for(x=0; x<16; x++)
2094
                h->dequant4_coeff[i][0][x] = 1<<6;
2095
        if(h->pps.transform_8x8_mode)
2096
            for(i=0; i<2; i++)
2097
                for(x=0; x<64; x++)
2098
                    h->dequant8_coeff[i][0][x] = 1<<6;
2099
    }
2100
}
2101

    
2102

    
2103
/**
2104
 * allocates tables.
2105
 * needs width/height
2106
 */
2107
static int alloc_tables(H264Context *h){
2108
    MpegEncContext * const s = &h->s;
2109
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2110
    int x,y;
2111

    
2112
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2113

    
2114
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2115
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
2116
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2117

    
2118
    if( h->pps.cabac ) {
2119
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2120
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2121
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2122
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2123
    }
2124

    
2125
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));
2126
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2127

    
2128
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2129
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2130
    for(y=0; y<s->mb_height; y++){
2131
        for(x=0; x<s->mb_width; x++){
2132
            const int mb_xy= x + y*s->mb_stride;
2133
            const int b_xy = 4*x + 4*y*h->b_stride;
2134
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2135

    
2136
            h->mb2b_xy [mb_xy]= b_xy;
2137
            h->mb2b8_xy[mb_xy]= b8_xy;
2138
        }
2139
    }
2140

    
2141
    s->obmc_scratchpad = NULL;
2142

    
2143
    if(!h->dequant4_coeff[0])
2144
        init_dequant_tables(h);
2145

    
2146
    return 0;
2147
fail:
2148
    free_tables(h);
2149
    return -1;
2150
}
2151

    
2152
/**
2153
 * Mimic alloc_tables(), but for every context thread.
2154
 */
2155
static void clone_tables(H264Context *dst, H264Context *src){
2156
    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
2157
    dst->non_zero_count           = src->non_zero_count;
2158
    dst->slice_table              = src->slice_table;
2159
    dst->cbp_table                = src->cbp_table;
2160
    dst->mb2b_xy                  = src->mb2b_xy;
2161
    dst->mb2b8_xy                 = src->mb2b8_xy;
2162
    dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
2163
    dst->mvd_table[0]             = src->mvd_table[0];
2164
    dst->mvd_table[1]             = src->mvd_table[1];
2165
    dst->direct_table             = src->direct_table;
2166

    
2167
    dst->s.obmc_scratchpad = NULL;
2168
    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2169
}
2170

    
2171
/**
2172
 * Init context
2173
 * Allocate buffers which are not shared amongst multiple threads.
2174
 */
2175
static int context_init(H264Context *h){
2176
    MpegEncContext * const s = &h->s;
2177

    
2178
    CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2179
    CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2180

    
2181
    // edge emu needs blocksize + filter length - 1 (=17x17 for halfpel / 21x21 for h264)
2182
    CHECKED_ALLOCZ(s->allocated_edge_emu_buffer,
2183
                   (s->width+64)*2*21*2); //(width + edge + align)*interlaced*MBsize*tolerance
2184
    s->edge_emu_buffer= s->allocated_edge_emu_buffer + (s->width+64)*2*21;
2185
    return 0;
2186
fail:
2187
    return -1; // free_tables will clean up for us
2188
}
2189

    
2190
static void common_init(H264Context *h){
2191
    MpegEncContext * const s = &h->s;
2192

    
2193
    s->width = s->avctx->width;
2194
    s->height = s->avctx->height;
2195
    s->codec_id= s->avctx->codec->id;
2196

    
2197
    ff_h264_pred_init(&h->hpc, s->codec_id);
2198

    
2199
    h->dequant_coeff_pps= -1;
2200
    s->unrestricted_mv=1;
2201
    s->decode=1; //FIXME
2202

    
2203
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2204
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2205
}
2206

    
2207
static int decode_init(AVCodecContext *avctx){
2208
    H264Context *h= avctx->priv_data;
2209
    MpegEncContext * const s = &h->s;
2210

    
2211
    MPV_decode_defaults(s);
2212

    
2213
    s->avctx = avctx;
2214
    common_init(h);
2215

    
2216
    s->out_format = FMT_H264;
2217
    s->workaround_bugs= avctx->workaround_bugs;
2218

    
2219
    // set defaults
2220
//    s->decode_mb= ff_h263_decode_mb;
2221
    s->quarter_sample = 1;
2222
    s->low_delay= 1;
2223
    avctx->pix_fmt= PIX_FMT_YUV420P;
2224

    
2225
    decode_init_vlc();
2226

    
2227
    if(avctx->extradata_size > 0 && avctx->extradata &&
2228
       *(char *)avctx->extradata == 1){
2229
        h->is_avc = 1;
2230
        h->got_avcC = 0;
2231
    } else {
2232
        h->is_avc = 0;
2233
    }
2234

    
2235
    h->thread_context[0] = h;
2236
    return 0;
2237
}
2238

    
2239
static int frame_start(H264Context *h){
2240
    MpegEncContext * const s = &h->s;
2241
    int i;
2242

    
2243
    if(MPV_frame_start(s, s->avctx) < 0)
2244
        return -1;
2245
    ff_er_frame_start(s);
2246

    
2247
    assert(s->linesize && s->uvlinesize);
2248

    
2249
    for(i=0; i<16; i++){
2250
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2251
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2252
    }
2253
    for(i=0; i<4; i++){
2254
        h->block_offset[16+i]=
2255
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2256
        h->block_offset[24+16+i]=
2257
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2258
    }
2259

    
2260
    /* can't be in alloc_tables because linesize isn't known there.
2261
     * FIXME: redo bipred weight to not require extra buffer? */
2262
    for(i = 0; i < s->avctx->thread_count; i++)
2263
        if(!h->thread_context[i]->s.obmc_scratchpad)
2264
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2265

    
2266
    /* some macroblocks will be accessed before they're available */
2267
    if(FRAME_MBAFF || s->avctx->thread_count > 1)
2268
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
2269

    
2270
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2271
    return 0;
2272
}
2273

    
2274
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){
2275
    MpegEncContext * const s = &h->s;
2276
    int i;
2277

    
2278
    src_y  -=   linesize;
2279
    src_cb -= uvlinesize;
2280
    src_cr -= uvlinesize;
2281

    
2282
    // There are two lines saved, the line above the the top macroblock of a pair,
2283
    // and the line above the bottom macroblock
2284
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
2285
    for(i=1; i<17; i++){
2286
        h->left_border[i]= src_y[15+i*  linesize];
2287
    }
2288

    
2289
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2290
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2291

    
2292
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2293
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
2294
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
2295
        for(i=1; i<9; i++){
2296
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
2297
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2298
        }
2299
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2300
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2301
    }
2302
}
2303

    
2304
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){
2305
    MpegEncContext * const s = &h->s;
2306
    int temp8, i;
2307
    uint64_t temp64;
2308
    int deblock_left;
2309
    int deblock_top;
2310
    int mb_xy;
2311

    
2312
    if(h->deblocking_filter == 2) {
2313
        mb_xy = s->mb_x + s->mb_y*s->mb_stride;
2314
        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2315
        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2316
    } else {
2317
        deblock_left = (s->mb_x > 0);
2318
        deblock_top =  (s->mb_y > 0);
2319
    }
2320

    
2321
    src_y  -=   linesize + 1;
2322
    src_cb -= uvlinesize + 1;
2323
    src_cr -= uvlinesize + 1;
2324

    
2325
#define XCHG(a,b,t,xchg)\
2326
t= a;\
2327
if(xchg)\
2328
    a= b;\
2329
b= t;
2330

    
2331
    if(deblock_left){
2332
        for(i = !deblock_top; i<17; i++){
2333
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2334
        }
2335
    }
2336

    
2337
    if(deblock_top){
2338
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2339
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2340
        if(s->mb_x+1 < s->mb_width){
2341
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2342
        }
2343
    }
2344

    
2345
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2346
        if(deblock_left){
2347
            for(i = !deblock_top; i<9; i++){
2348
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
2349
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2350
            }
2351
        }
2352
        if(deblock_top){
2353
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2354
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2355
        }
2356
    }
2357
}
2358

    
2359
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2360
    MpegEncContext * const s = &h->s;
2361
    int i;
2362

    
2363
    src_y  -= 2 *   linesize;
2364
    src_cb -= 2 * uvlinesize;
2365
    src_cr -= 2 * uvlinesize;
2366

    
2367
    // There are two lines saved, the line above the the top macroblock of a pair,
2368
    // and the line above the bottom macroblock
2369
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
2370
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
2371
    for(i=2; i<34; i++){
2372
        h->left_border[i]= src_y[15+i*  linesize];
2373
    }
2374

    
2375
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
2376
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
2377
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
2378
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
2379

    
2380
    if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2381
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
2382
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
2383
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
2384
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
2385
        for(i=2; i<18; i++){
2386
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
2387
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
2388
        }
2389
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
2390
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
2391
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
2392
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
2393
    }
2394
}
2395

    
2396
static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
2397
    MpegEncContext * const s = &h->s;
2398
    int temp8, i;
2399
    uint64_t temp64;
2400
    int deblock_left = (s->mb_x > 0);
2401
    int deblock_top  = (s->mb_y > 1);
2402

    
2403
    tprintf(s->avctx, "xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
2404

    
2405
    src_y  -= 2 *   linesize + 1;
2406
    src_cb -= 2 * uvlinesize + 1;
2407
    src_cr -= 2 * uvlinesize + 1;
2408

    
2409
#define XCHG(a,b,t,xchg)\
2410
t= a;\
2411
if(xchg)\
2412
    a= b;\
2413
b= t;
2414

    
2415
    if(deblock_left){
2416
        for(i = (!deblock_top)<<1; i<34; i++){
2417
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2418
        }
2419
    }
2420

    
2421
    if(deblock_top){
2422
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2423
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2424
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
2425
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
2426
        if(s->mb_x+1 < s->mb_width){
2427
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2428
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
2429
        }
2430
    }
2431

    
2432
    if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2433
        if(deblock_left){
2434
            for(i = (!deblock_top) << 1; i<18; i++){
2435
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
2436
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
2437
            }
2438
        }
2439
        if(deblock_top){
2440
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2441
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2442
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
2443
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
2444
        }
2445
    }
2446
}
2447

    
2448
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2449
    MpegEncContext * const s = &h->s;
2450
    const int mb_x= s->mb_x;
2451
    const int mb_y= s->mb_y;
2452
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2453
    const int mb_type= s->current_picture.mb_type[mb_xy];
2454
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2455
    int linesize, uvlinesize /*dct_offset*/;
2456
    int i;
2457
    int *block_offset = &h->block_offset[0];
2458
    const unsigned int bottom = mb_y & 1;
2459
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass), is_h264 = (simple || s->codec_id == CODEC_ID_H264);
2460
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2461
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2462

    
2463
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2464
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2465
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2466

    
2467
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2468
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2469

    
2470
    if (!simple && MB_FIELD) {
2471
        linesize   = h->mb_linesize   = s->linesize * 2;
2472
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2473
        block_offset = &h->block_offset[24];
2474
        if(mb_y&1){ //FIXME move out of this func?
2475
            dest_y -= s->linesize*15;
2476
            dest_cb-= s->uvlinesize*7;
2477
            dest_cr-= s->uvlinesize*7;
2478
        }
2479
        if(FRAME_MBAFF) {
2480
            int list;
2481
            for(list=0; list<h->list_count; list++){
2482
                if(!USES_LIST(mb_type, list))
2483
                    continue;
2484
                if(IS_16X16(mb_type)){
2485
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
2486
                    fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1);
2487
                }else{
2488
                    for(i=0; i<16; i+=4){
2489
                        //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
2490
                        int ref = h->ref_cache[list][scan8[i]];
2491
                        if(ref >= 0)
2492
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1);
2493
                    }
2494
                }
2495
            }
2496
        }
2497
    } else {
2498
        linesize   = h->mb_linesize   = s->linesize;
2499
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2500
//        dct_offset = s->linesize * 16;
2501
    }
2502

    
2503
    if(transform_bypass){
2504
        idct_dc_add =
2505
        idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2506
    }else if(IS_8x8DCT(mb_type)){
2507
        idct_dc_add = s->dsp.h264_idct8_dc_add;
2508
        idct_add = s->dsp.h264_idct8_add;
2509
    }else{
2510
        idct_dc_add = s->dsp.h264_idct_dc_add;
2511
        idct_add = s->dsp.h264_idct_add;
2512
    }
2513

    
2514
    if(!simple && FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
2515
       && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
2516
        int mbt_y = mb_y&~1;
2517
        uint8_t *top_y  = s->current_picture.data[0] + (mbt_y * 16* s->linesize  ) + mb_x * 16;
2518
        uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
2519
        uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
2520
        xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
2521
    }
2522

    
2523
    if (!simple && IS_INTRA_PCM(mb_type)) {
2524
        unsigned int x, y;
2525

    
2526
        // The pixels are stored in h->mb array in the same order as levels,
2527
        // copy them in output in the correct order.
2528
        for(i=0; i<16; i++) {
2529
            for (y=0; y<4; y++) {
2530
                for (x=0; x<4; x++) {
2531
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
2532
                }
2533
            }
2534
        }
2535
        for(i=16; i<16+4; i++) {
2536
            for (y=0; y<4; y++) {
2537
                for (x=0; x<4; x++) {
2538
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2539
                }
2540
            }
2541
        }
2542
        for(i=20; i<20+4; i++) {
2543
            for (y=0; y<4; y++) {
2544
                for (x=0; x<4; x++) {
2545
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2546
                }
2547
            }
2548
        }
2549
    } else {
2550
        if(IS_INTRA(mb_type)){
2551
            if(h->deblocking_filter && (simple || !FRAME_MBAFF))
2552
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2553

    
2554
            if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2555
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2556
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2557
            }
2558

    
2559
            if(IS_INTRA4x4(mb_type)){
2560
                if(simple || !s->encoding){
2561
                    if(IS_8x8DCT(mb_type)){
2562
                        for(i=0; i<16; i+=4){
2563
                            uint8_t * const ptr= dest_y + block_offset[i];
2564
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2565
                            const int nnz = h->non_zero_count_cache[ scan8[i] ];
2566
                            h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2567
                                                   (h->topright_samples_available<<i)&0x4000, linesize);
2568
                            if(nnz){
2569
                                if(nnz == 1 && h->mb[i*16])
2570
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
2571
                                else
2572
                                    idct_add(ptr, h->mb + i*16, linesize);
2573
                            }
2574
                        }
2575
                    }else
2576
                    for(i=0; i<16; i++){
2577
                        uint8_t * const ptr= dest_y + block_offset[i];
2578
                        uint8_t *topright;
2579
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2580
                        int nnz, tr;
2581

    
2582
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2583
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2584
                            assert(mb_y || linesize <= block_offset[i]);
2585
                            if(!topright_avail){
2586
                                tr= ptr[3 - linesize]*0x01010101;
2587
                                topright= (uint8_t*) &tr;
2588
                            }else
2589
                                topright= ptr + 4 - linesize;
2590
                        }else
2591
                            topright= NULL;
2592

    
2593
                        h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2594
                        nnz = h->non_zero_count_cache[ scan8[i] ];
2595
                        if(nnz){
2596
                            if(is_h264){
2597
                                if(nnz == 1 && h->mb[i*16])
2598
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
2599
                                else
2600
                                    idct_add(ptr, h->mb + i*16, linesize);
2601
                            }else
2602
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2603
                        }
2604
                    }
2605
                }
2606
            }else{
2607
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2608
                if(is_h264){
2609
                    if(!transform_bypass)
2610
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
2611
                }else
2612
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2613
            }
2614
            if(h->deblocking_filter && (simple || !FRAME_MBAFF))
2615
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2616
        }else if(is_h264){
2617
            hl_motion(h, dest_y, dest_cb, dest_cr,
2618
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2619
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2620
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2621
        }
2622

    
2623

    
2624
        if(!IS_INTRA4x4(mb_type)){
2625
            if(is_h264){
2626
                if(IS_INTRA16x16(mb_type)){
2627
                    for(i=0; i<16; i++){
2628
                        if(h->non_zero_count_cache[ scan8[i] ])
2629
                            idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2630
                        else if(h->mb[i*16])
2631
                            idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2632
                    }
2633
                }else{
2634
                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2635
                    for(i=0; i<16; i+=di){
2636
                        int nnz = h->non_zero_count_cache[ scan8[i] ];
2637
                        if(nnz){
2638
                            if(nnz==1 && h->mb[i*16])
2639
                                idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2640
                            else
2641
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2642
                        }
2643
                    }
2644
                }
2645
            }else{
2646
                for(i=0; i<16; i++){
2647
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2648
                        uint8_t * const ptr= dest_y + block_offset[i];
2649
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2650
                    }
2651
                }
2652
            }
2653
        }
2654

    
2655
        if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2656
            uint8_t *dest[2] = {dest_cb, dest_cr};
2657
            if(transform_bypass){
2658
                idct_add = idct_dc_add = s->dsp.add_pixels4;
2659
            }else{
2660
                idct_add = s->dsp.h264_idct_add;
2661
                idct_dc_add = s->dsp.h264_idct_dc_add;
2662
                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]);
2663
                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]);
2664
            }
2665
            if(is_h264){
2666
                for(i=16; i<16+8; i++){
2667
                    if(h->non_zero_count_cache[ scan8[i] ])
2668
                        idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2669
                    else if(h->mb[i*16])
2670
                        idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2671
                }
2672
            }else{
2673
                for(i=16; i<16+8; i++){
2674
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2675
                        uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
2676
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2677
                    }
2678
                }
2679
            }
2680
        }
2681
    }
2682
    if(h->deblocking_filter) {
2683
        if (!simple && FRAME_MBAFF) {
2684
            //FIXME try deblocking one mb at a time?
2685
            // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
2686
            const int mb_y = s->mb_y - 1;
2687
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
2688
            const int mb_xy= mb_x + mb_y*s->mb_stride;
2689
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
2690
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
2691
            if (!bottom) return;
2692
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2693
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2694
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2695

    
2696
            if(IS_INTRA(mb_type_top | mb_type_bottom))
2697
                xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
2698

    
2699
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
2700
            // deblock a pair
2701
            // top
2702
            s->mb_y--;
2703
            tprintf(h->s.avctx, "call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
2704
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
2705
            h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2706
            h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2707
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
2708
            // bottom
2709
            s->mb_y++;
2710
            tprintf(h->s.avctx, "call mbaff filter_mb\n");
2711
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
2712
            h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
2713
            h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
2714
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2715
        } else {
2716
            tprintf(h->s.avctx, "call filter_mb\n");
2717
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
2718
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
2719
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2720
        }
2721
    }
2722
}
2723

    
2724
/**
2725
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2726
 */
2727
static void hl_decode_mb_simple(H264Context *h){
2728
    hl_decode_mb_internal(h, 1);
2729
}
2730

    
2731
/**
2732
 * Process a macroblock; this handles edge cases, such as interlacing.
2733
 */
2734
static void av_noinline hl_decode_mb_complex(H264Context *h){
2735
    hl_decode_mb_internal(h, 0);
2736
}
2737

    
2738
static void hl_decode_mb(H264Context *h){
2739
    MpegEncContext * const s = &h->s;
2740
    const int mb_x= s->mb_x;
2741
    const int mb_y= s->mb_y;
2742
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2743
    const int mb_type= s->current_picture.mb_type[mb_xy];
2744
    int is_complex = FRAME_MBAFF || MB_FIELD || IS_INTRA_PCM(mb_type) || s->codec_id != CODEC_ID_H264 || (ENABLE_GRAY && (s->flags&CODEC_FLAG_GRAY)) || s->encoding;
2745

    
2746
    if(!s->decode)
2747
        return;
2748

    
2749
    if (is_complex)
2750
        hl_decode_mb_complex(h);
2751
    else hl_decode_mb_simple(h);
2752
}
2753

    
2754
/**
2755
 * fills the default_ref_list.
2756
 */
2757
static int fill_default_ref_list(H264Context *h){
2758
    MpegEncContext * const s = &h->s;
2759
    int i;
2760
    int smallest_poc_greater_than_current = -1;
2761
    Picture sorted_short_ref[32];
2762

    
2763
    if(h->slice_type==B_TYPE){
2764
        int out_i;
2765
        int limit= INT_MIN;
2766

    
2767
        /* sort frame according to poc in B slice */
2768
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2769
            int best_i=INT_MIN;
2770
            int best_poc=INT_MAX;
2771

    
2772
            for(i=0; i<h->short_ref_count; i++){
2773
                const int poc= h->short_ref[i]->poc;
2774
                if(poc > limit && poc < best_poc){
2775
                    best_poc= poc;
2776
                    best_i= i;
2777
                }
2778
            }
2779

    
2780
            assert(best_i != INT_MIN);
2781

    
2782
            limit= best_poc;
2783
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2784
            tprintf(h->s.avctx, "sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num);
2785
            if (-1 == smallest_poc_greater_than_current) {
2786
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
2787
                    smallest_poc_greater_than_current = out_i;
2788
                }
2789
            }
2790
        }
2791
    }
2792

    
2793
    if(s->picture_structure == PICT_FRAME){
2794
        if(h->slice_type==B_TYPE){
2795
            int list;
2796
            tprintf(h->s.avctx, "current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
2797

    
2798
            // find the largest poc
2799
            for(list=0; list<2; list++){
2800
                int index = 0;
2801
                int j= -99;
2802
                int step= list ? -1 : 1;
2803

    
2804
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
2805
                    while(j<0 || j>= h->short_ref_count){
2806
                        if(j != -99 && step == (list ? -1 : 1))
2807
                            return -1;
2808
                        step = -step;
2809
                        j= smallest_poc_greater_than_current + (step>>1);
2810
                    }
2811
                    if(sorted_short_ref[j].reference != 3) continue;
2812
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
2813
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
2814
                }
2815

    
2816
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
2817
                    if(h->long_ref[i] == NULL) continue;
2818
                    if(h->long_ref[i]->reference != 3) continue;
2819

    
2820
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2821
                    h->default_ref_list[ list ][index++].pic_id= i;;
2822
                }
2823

    
2824
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
2825
                    // swap the two first elements of L1 when
2826
                    // L0 and L1 are identical
2827
                    Picture temp= h->default_ref_list[1][0];
2828
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
2829
                    h->default_ref_list[1][1] = temp;
2830
                }
2831

    
2832
                if(index < h->ref_count[ list ])
2833
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2834
            }
2835
        }else{
2836
            int index=0;
2837
            for(i=0; i<h->short_ref_count; i++){
2838
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2839
                h->default_ref_list[0][index  ]= *h->short_ref[i];
2840
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2841
            }
2842
            for(i = 0; i < 16; i++){
2843
                if(h->long_ref[i] == NULL) continue;
2844
                if(h->long_ref[i]->reference != 3) continue;
2845
                h->default_ref_list[0][index  ]= *h->long_ref[i];
2846
                h->default_ref_list[0][index++].pic_id= i;;
2847
            }
2848
            if(index < h->ref_count[0])
2849
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2850
        }
2851
    }else{ //FIELD
2852
        if(h->slice_type==B_TYPE){
2853
        }else{
2854
            //FIXME second field balh
2855
        }
2856
    }
2857
#ifdef TRACE
2858
    for (i=0; i<h->ref_count[0]; i++) {
2859
        tprintf(h->s.avctx, "List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]);
2860
    }
2861
    if(h->slice_type==B_TYPE){
2862
        for (i=0; i<h->ref_count[1]; i++) {
2863
            tprintf(h->s.avctx, "List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[0][i].data[0]);
2864
        }
2865
    }
2866
#endif
2867
    return 0;
2868
}
2869

    
2870
static void print_short_term(H264Context *h);
2871
static void print_long_term(H264Context *h);
2872

    
2873
static int decode_ref_pic_list_reordering(H264Context *h){
2874
    MpegEncContext * const s = &h->s;
2875
    int list, index;
2876

    
2877
    print_short_term(h);
2878
    print_long_term(h);
2879
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
2880

    
2881
    for(list=0; list<h->list_count; list++){
2882
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2883

    
2884
        if(get_bits1(&s->gb)){
2885
            int pred= h->curr_pic_num;
2886

    
2887
            for(index=0; ; index++){
2888
                unsigned int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2889
                unsigned int pic_id;
2890
                int i;
2891
                Picture *ref = NULL;
2892

    
2893
                if(reordering_of_pic_nums_idc==3)
2894
                    break;
2895

    
2896
                if(index >= h->ref_count[list]){
2897
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2898
                    return -1;
2899
                }
2900

    
2901
                if(reordering_of_pic_nums_idc<3){
2902
                    if(reordering_of_pic_nums_idc<2){
2903
                        const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2904

    
2905
                        if(abs_diff_pic_num >= h->max_pic_num){
2906
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2907
                            return -1;
2908
                        }
2909

    
2910
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2911
                        else                                pred+= abs_diff_pic_num;
2912
                        pred &= h->max_pic_num - 1;
2913

    
2914
                        for(i= h->short_ref_count-1; i>=0; i--){
2915
                            ref = h->short_ref[i];
2916
                            assert(ref->reference == 3);
2917
                            assert(!ref->long_ref);
2918
                            if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
2919
                                break;
2920
                        }
2921
                        if(i>=0)
2922
                            ref->pic_id= ref->frame_num;
2923
                    }else{
2924
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2925
                        if(pic_id>31){
2926
                            av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
2927
                            return -1;
2928
                        }
2929
                        ref = h->long_ref[pic_id];
2930
                        if(ref){
2931
                            ref->pic_id= pic_id;
2932
                            assert(ref->reference == 3);
2933
                            assert(ref->long_ref);
2934
                            i=0;
2935
                        }else{
2936
                            i=-1;
2937
                        }
2938
                    }
2939

    
2940
                    if (i < 0) {
2941
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2942
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2943
                    } else {
2944
                        for(i=index; i+1<h->ref_count[list]; i++){
2945
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
2946
                                break;
2947
                        }
2948
                        for(; i > index; i--){
2949
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2950
                        }
2951
                        h->ref_list[list][index]= *ref;
2952
                    }
2953
                }else{
2954
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2955
                    return -1;
2956
                }
2957
            }
2958
        }
2959
    }
2960
    for(list=0; list<h->list_count; list++){
2961
        for(index= 0; index < h->ref_count[list]; index++){
2962
            if(!h->ref_list[list][index].data[0])
2963
                h->ref_list[list][index]= s->current_picture;
2964
        }
2965
    }
2966

    
2967
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
2968
        direct_dist_scale_factor(h);
2969
    direct_ref_list_init(h);
2970
    return 0;
2971
}
2972

    
2973
static void fill_mbaff_ref_list(H264Context *h){
2974
    int list, i, j;
2975
    for(list=0; list<2; list++){ //FIXME try list_count
2976
        for(i=0; i<h->ref_count[list]; i++){
2977
            Picture *frame = &h->ref_list[list][i];
2978
            Picture *field = &h->ref_list[list][16+2*i];
2979
            field[0] = *frame;
2980
            for(j=0; j<3; j++)
2981
                field[0].linesize[j] <<= 1;
2982
            field[1] = field[0];
2983
            for(j=0; j<3; j++)
2984
                field[1].data[j] += frame->linesize[j];
2985

    
2986
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
2987
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
2988
            for(j=0; j<2; j++){
2989
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
2990
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
2991
            }
2992
        }
2993
    }
2994
    for(j=0; j<h->ref_count[1]; j++){
2995
        for(i=0; i<h->ref_count[0]; i++)
2996
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
2997
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
2998
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
2999
    }
3000
}
3001

    
3002
static int pred_weight_table(H264Context *h){
3003
    MpegEncContext * const s = &h->s;
3004
    int list, i;
3005
    int luma_def, chroma_def;
3006

    
3007
    h->use_weight= 0;
3008
    h->use_weight_chroma= 0;
3009
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3010
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3011
    luma_def = 1<<h->luma_log2_weight_denom;
3012
    chroma_def = 1<<h->chroma_log2_weight_denom;
3013

    
3014
    for(list=0; list<2; list++){
3015
        for(i=0; i<h->ref_count[list]; i++){
3016
            int luma_weight_flag, chroma_weight_flag;
3017

    
3018
            luma_weight_flag= get_bits1(&s->gb);
3019
            if(luma_weight_flag){
3020
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3021
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3022
                if(   h->luma_weight[list][i] != luma_def
3023
                   || h->luma_offset[list][i] != 0)
3024
                    h->use_weight= 1;
3025
            }else{
3026
                h->luma_weight[list][i]= luma_def;
3027
                h->luma_offset[list][i]= 0;
3028
            }
3029

    
3030
            chroma_weight_flag= get_bits1(&s->gb);
3031
            if(chroma_weight_flag){
3032
                int j;
3033
                for(j=0; j<2; j++){
3034
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3035
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3036
                    if(   h->chroma_weight[list][i][j] != chroma_def
3037
                       || h->chroma_offset[list][i][j] != 0)
3038
                        h->use_weight_chroma= 1;
3039
                }
3040
            }else{
3041
                int j;
3042
                for(j=0; j<2; j++){
3043
                    h->chroma_weight[list][i][j]= chroma_def;
3044
                    h->chroma_offset[list][i][j]= 0;
3045
                }
3046
            }
3047
        }
3048
        if(h->slice_type != B_TYPE) break;
3049
    }
3050
    h->use_weight= h->use_weight || h->use_weight_chroma;
3051
    return 0;
3052
}
3053

    
3054
static void implicit_weight_table(H264Context *h){
3055
    MpegEncContext * const s = &h->s;
3056
    int ref0, ref1;
3057
    int cur_poc = s->current_picture_ptr->poc;
3058

    
3059
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3060
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3061
        h->use_weight= 0;
3062
        h->use_weight_chroma= 0;
3063
        return;
3064
    }
3065

    
3066
    h->use_weight= 2;
3067
    h->use_weight_chroma= 2;
3068
    h->luma_log2_weight_denom= 5;
3069
    h->chroma_log2_weight_denom= 5;
3070

    
3071
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3072
        int poc0 = h->ref_list[0][ref0].poc;
3073
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3074
            int poc1 = h->ref_list[1][ref1].poc;
3075
            int td = av_clip(poc1 - poc0, -128, 127);
3076
            if(td){
3077
                int tb = av_clip(cur_poc - poc0, -128, 127);
3078
                int tx = (16384 + (FFABS(td) >> 1)) / td;
3079
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3080
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3081
                    h->implicit_weight[ref0][ref1] = 32;
3082
                else
3083
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3084
            }else
3085
                h->implicit_weight[ref0][ref1] = 32;
3086
        }
3087
    }
3088
}
3089

    
3090
static inline void unreference_pic(H264Context *h, Picture *pic){
3091
    int i;
3092
    pic->reference=0;
3093
    if(pic == h->delayed_output_pic)
3094
        pic->reference=1;
3095
    else{
3096
        for(i = 0; h->delayed_pic[i]; i++)
3097
            if(pic == h->delayed_pic[i]){
3098
                pic->reference=1;
3099
                break;
3100
            }
3101
    }
3102
}
3103

    
3104
/**
3105
 * instantaneous decoder refresh.
3106
 */
3107
static void idr(H264Context *h){
3108
    int i;
3109

    
3110
    for(i=0; i<16; i++){
3111
        if (h->long_ref[i] != NULL) {
3112
            unreference_pic(h, h->long_ref[i]);
3113
            h->long_ref[i]= NULL;
3114
        }
3115
    }
3116
    h->long_ref_count=0;
3117

    
3118
    for(i=0; i<h->short_ref_count; i++){
3119
        unreference_pic(h, h->short_ref[i]);
3120
        h->short_ref[i]= NULL;
3121
    }
3122
    h->short_ref_count=0;
3123
}
3124

    
3125
/* forget old pics after a seek */
3126
static void flush_dpb(AVCodecContext *avctx){
3127
    H264Context *h= avctx->priv_data;
3128
    int i;
3129
    for(i=0; i<16; i++) {
3130
        if(h->delayed_pic[i])
3131
            h->delayed_pic[i]->reference= 0;
3132
        h->delayed_pic[i]= NULL;
3133
    }
3134
    if(h->delayed_output_pic)
3135
        h->delayed_output_pic->reference= 0;
3136
    h->delayed_output_pic= NULL;
3137
    idr(h);
3138
    if(h->s.current_picture_ptr)
3139
        h->s.current_picture_ptr->reference= 0;
3140
}
3141

    
3142
/**
3143
 *
3144
 * @return the removed picture or NULL if an error occurs
3145
 */
3146
static Picture * remove_short(H264Context *h, int frame_num){
3147
    MpegEncContext * const s = &h->s;
3148
    int i;
3149

    
3150
    if(s->avctx->debug&FF_DEBUG_MMCO)
3151
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3152

    
3153
    for(i=0; i<h->short_ref_count; i++){
3154
        Picture *pic= h->short_ref[i];
3155
        if(s->avctx->debug&FF_DEBUG_MMCO)
3156
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3157
        if(pic->frame_num == frame_num){
3158
            h->short_ref[i]= NULL;
3159
            if (--h->short_ref_count)
3160
                memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i)*sizeof(Picture*));
3161
            return pic;
3162
        }
3163
    }
3164
    return NULL;
3165
}
3166

    
3167
/**
3168
 *
3169
 * @return the removed picture or NULL if an error occurs
3170
 */
3171
static Picture * remove_long(H264Context *h, int i){
3172
    Picture *pic;
3173

    
3174
    pic= h->long_ref[i];
3175
    h->long_ref[i]= NULL;
3176
    if(pic) h->long_ref_count--;
3177

    
3178
    return pic;
3179
}
3180

    
3181
/**
3182
 * print short term list
3183
 */
3184
static void print_short_term(H264Context *h) {
3185
    uint32_t i;
3186
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3187
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3188
        for(i=0; i<h->short_ref_count; i++){
3189
            Picture *pic= h->short_ref[i];
3190
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3191
        }
3192
    }
3193
}
3194

    
3195
/**
3196
 * print long term list
3197
 */
3198
static void print_long_term(H264Context *h) {
3199
    uint32_t i;
3200
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3201
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3202
        for(i = 0; i < 16; i++){
3203
            Picture *pic= h->long_ref[i];
3204
            if (pic) {
3205
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3206
            }
3207
        }
3208
    }
3209
}
3210

    
3211
/**
3212
 * Executes the reference picture marking (memory management control operations).
3213
 */
3214
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3215
    MpegEncContext * const s = &h->s;
3216
    int i, j;
3217
    int current_is_long=0;
3218
    Picture *pic;
3219

    
3220
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3221
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3222

    
3223
    for(i=0; i<mmco_count; i++){
3224
        if(s->avctx->debug&FF_DEBUG_MMCO)
3225
            av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
3226

    
3227
        switch(mmco[i].opcode){
3228
        case MMCO_SHORT2UNUSED:
3229
            pic= remove_short(h, mmco[i].short_frame_num);
3230
            if(pic)
3231
                unreference_pic(h, pic);
3232
            else if(s->avctx->debug&FF_DEBUG_MMCO)
3233
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_short() failure\n");
3234
            break;
3235
        case MMCO_SHORT2LONG:
3236
            pic= remove_long(h, mmco[i].long_index);
3237
            if(pic) unreference_pic(h, pic);
3238

    
3239
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3240
            if (h->long_ref[ mmco[i].long_index ]){
3241
                h->long_ref[ mmco[i].long_index ]->long_ref=1;
3242
                h->long_ref_count++;
3243
            }
3244
            break;
3245
        case MMCO_LONG2UNUSED:
3246
            pic= remove_long(h, mmco[i].long_index);
3247
            if(pic)
3248
                unreference_pic(h, pic);
3249
            else if(s->avctx->debug&FF_DEBUG_MMCO)
3250
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_long() failure\n");
3251
            break;
3252
        case MMCO_LONG:
3253
            pic= remove_long(h, mmco[i].long_index);
3254
            if(pic) unreference_pic(h, pic);
3255

    
3256
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3257
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3258
            h->long_ref_count++;
3259

    
3260
            current_is_long=1;
3261
            break;
3262
        case MMCO_SET_MAX_LONG:
3263
            assert(mmco[i].long_index <= 16);
3264
            // just remove the long term which index is greater than new max
3265
            for(j = mmco[i].long_index; j<16; j++){
3266
                pic = remove_long(h, j);
3267
                if (pic) unreference_pic(h, pic);
3268
            }
3269
            break;
3270
        case MMCO_RESET:
3271
            while(h->short_ref_count){
3272
                pic= remove_short(h, h->short_ref[0]->frame_num);
3273
                if(pic) unreference_pic(h, pic);
3274
            }
3275
            for(j = 0; j < 16; j++) {
3276
                pic= remove_long(h, j);
3277
                if(pic) unreference_pic(h, pic);
3278
            }
3279
            break;
3280
        default: assert(0);
3281
        }
3282
    }
3283

    
3284
    if(!current_is_long){
3285
        pic= remove_short(h, s->current_picture_ptr->frame_num);
3286
        if(pic){
3287
            unreference_pic(h, pic);
3288
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3289
        }
3290

    
3291
        if(h->short_ref_count)
3292
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3293

    
3294
        h->short_ref[0]= s->current_picture_ptr;
3295
        h->short_ref[0]->long_ref=0;
3296
        h->short_ref_count++;
3297
    }
3298

    
3299
    print_short_term(h);
3300
    print_long_term(h);
3301
    return 0;
3302
}
3303

    
3304
static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
3305
    MpegEncContext * const s = &h->s;
3306
    int i;
3307

    
3308
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3309
        s->broken_link= get_bits1(gb) -1;
3310
        h->mmco[0].long_index= get_bits1(gb) - 1; // current_long_term_idx
3311
        if(h->mmco[0].long_index == -1)
3312
            h->mmco_index= 0;
3313
        else{
3314
            h->mmco[0].opcode= MMCO_LONG;
3315
            h->mmco_index= 1;
3316
        }
3317
    }else{
3318
        if(get_bits1(gb)){ // adaptive_ref_pic_marking_mode_flag
3319
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
3320
                MMCOOpcode opcode= get_ue_golomb(gb);
3321

    
3322
                h->mmco[i].opcode= opcode;
3323
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3324
                    h->mmco[i].short_frame_num= (h->frame_num - get_ue_golomb(gb) - 1) & ((1<<h->sps.log2_max_frame_num)-1); //FIXME fields
3325
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3326
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
3327
                        return -1;
3328
                    }*/
3329
                }
3330
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3331
                    unsigned int long_index= get_ue_golomb(gb);
3332
                    if(/*h->mmco[i].long_index >= h->long_ref_count || h->long_ref[ h->mmco[i].long_index ] == NULL*/ long_index >= 16){
3333
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3334
                        return -1;
3335
                    }
3336
                    h->mmco[i].long_index= long_index;
3337
                }
3338

    
3339
                if(opcode > (unsigned)MMCO_LONG){
3340
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3341
                    return -1;
3342
                }
3343
                if(opcode == MMCO_END)
3344
                    break;
3345
            }
3346
            h->mmco_index= i;
3347
        }else{
3348
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3349

    
3350
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3351
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3352
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3353
                h->mmco_index= 1;
3354
            }else
3355
                h->mmco_index= 0;
3356
        }
3357
    }
3358

    
3359
    return 0;
3360
}
3361

    
3362
static int init_poc(H264Context *h){
3363
    MpegEncContext * const s = &h->s;
3364
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3365
    int field_poc[2];
3366

    
3367
    if(h->nal_unit_type == NAL_IDR_SLICE){
3368
        h->frame_num_offset= 0;
3369
    }else{
3370
        if(h->frame_num < h->prev_frame_num)
3371
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3372
        else
3373
            h->frame_num_offset= h->prev_frame_num_offset;
3374
    }
3375

    
3376
    if(h->sps.poc_type==0){
3377
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3378

    
3379
        if(h->nal_unit_type == NAL_IDR_SLICE){
3380
             h->prev_poc_msb=
3381
             h->prev_poc_lsb= 0;
3382
        }
3383

    
3384
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3385
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3386
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3387
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3388
        else
3389
            h->poc_msb = h->prev_poc_msb;
3390
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3391
        field_poc[0] =
3392
        field_poc[1] = h->poc_msb + h->poc_lsb;
3393
        if(s->picture_structure == PICT_FRAME)
3394
            field_poc[1] += h->delta_poc_bottom;
3395
    }else if(h->sps.poc_type==1){
3396
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3397
        int i;
3398

    
3399
        if(h->sps.poc_cycle_length != 0)
3400
            abs_frame_num = h->frame_num_offset + h->frame_num;
3401
        else
3402
            abs_frame_num = 0;
3403

    
3404
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
3405
            abs_frame_num--;
3406

    
3407
        expected_delta_per_poc_cycle = 0;
3408
        for(i=0; i < h->sps.poc_cycle_length; i++)
3409
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3410

    
3411
        if(abs_frame_num > 0){
3412
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3413
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3414

    
3415
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3416
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
3417
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3418
        } else
3419
            expectedpoc = 0;
3420

    
3421
        if(h->nal_ref_idc == 0)
3422
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3423

    
3424
        field_poc[0] = expectedpoc + h->delta_poc[0];
3425
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3426

    
3427
        if(s->picture_structure == PICT_FRAME)
3428
            field_poc[1] += h->delta_poc[1];
3429
    }else{
3430
        int poc;
3431
        if(h->nal_unit_type == NAL_IDR_SLICE){
3432
            poc= 0;
3433
        }else{
3434
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3435
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3436
        }
3437
        field_poc[0]= poc;
3438
        field_poc[1]= poc;
3439
    }
3440

    
3441
    if(s->picture_structure != PICT_BOTTOM_FIELD)
3442
        s->current_picture_ptr->field_poc[0]= field_poc[0];
3443
    if(s->picture_structure != PICT_TOP_FIELD)
3444
        s->current_picture_ptr->field_poc[1]= field_poc[1];
3445
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3446
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3447

    
3448
    return 0;
3449
}
3450

    
3451

    
3452
/**
3453
 * initialize scan tables
3454
 */
3455
static void init_scan_tables(H264Context *h){
3456
    MpegEncContext * const s = &h->s;
3457
    int i;
3458
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
3459
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
3460
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
3461
    }else{
3462
        for(i=0; i<16; i++){
3463
#define T(x) (x>>2) | ((x<<2) & 0xF)
3464
            h->zigzag_scan[i] = T(zigzag_scan[i]);
3465
            h-> field_scan[i] = T( field_scan[i]);
3466
#undef T
3467
        }
3468
    }
3469
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
3470
        memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));
3471
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
3472
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
3473
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
3474
    }else{
3475
        for(i=0; i<64; i++){
3476
#define T(x) (x>>3) | ((x&7)<<3)
3477
            h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);
3478
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
3479
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
3480
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
3481
#undef T
3482
        }
3483
    }
3484
    if(h->sps.transform_bypass){ //FIXME same ugly
3485
        h->zigzag_scan_q0          = zigzag_scan;
3486
        h->zigzag_scan8x8_q0       = zigzag_scan8x8;
3487
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
3488
        h->field_scan_q0           = field_scan;
3489
        h->field_scan8x8_q0        = field_scan8x8;
3490
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
3491
    }else{
3492
        h->zigzag_scan_q0          = h->zigzag_scan;
3493
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
3494
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
3495
        h->field_scan_q0           = h->field_scan;
3496
        h->field_scan8x8_q0        = h->field_scan8x8;
3497
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
3498
    }
3499
}
3500

    
3501
/**
3502
 * Replicates H264 "master" context to thread contexts.
3503
 */
3504
static void clone_slice(H264Context *dst, H264Context *src)
3505
{
3506
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
3507
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
3508
    dst->s.current_picture      = src->s.current_picture;
3509
    dst->s.linesize             = src->s.linesize;
3510
    dst->s.uvlinesize           = src->s.uvlinesize;
3511

    
3512
    dst->prev_poc_msb           = src->prev_poc_msb;
3513
    dst->prev_poc_lsb           = src->prev_poc_lsb;
3514
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
3515
    dst->prev_frame_num         = src->prev_frame_num;
3516
    dst->short_ref_count        = src->short_ref_count;
3517

    
3518
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
3519
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
3520
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
3521
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
3522

    
3523
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
3524
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
3525
}
3526

    
3527
/**
3528
 * decodes a slice header.
3529
 * this will allso call MPV_common_init() and frame_start() as needed
3530
 *
3531
 * @param h h264context
3532
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
3533
 *
3534
 * @return 0 if okay, <0 if an error occured, 1 if decoding must not be multithreaded
3535
 */
3536
static int decode_slice_header(H264Context *h, H264Context *h0){
3537
    MpegEncContext * const s = &h->s;
3538
    unsigned int first_mb_in_slice;
3539
    unsigned int pps_id;
3540
    int num_ref_idx_active_override_flag;
3541
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3542
    unsigned int slice_type, tmp, i;
3543
    int default_ref_list_done = 0;
3544

    
3545
    s->current_picture.reference= h->nal_ref_idc != 0;
3546
    s->dropable= h->nal_ref_idc == 0;
3547

    
3548
    first_mb_in_slice= get_ue_golomb(&s->gb);
3549

    
3550
    if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
3551
        h0->current_slice = 0;
3552
        s->current_picture_ptr= NULL;
3553
    }
3554

    
3555
    slice_type= get_ue_golomb(&s->gb);
3556
    if(slice_type > 9){
3557
        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);
3558
        return -1;
3559
    }
3560
    if(slice_type > 4){
3561
        slice_type -= 5;
3562
        h->slice_type_fixed=1;
3563
    }else
3564
        h->slice_type_fixed=0;
3565

    
3566
    slice_type= slice_type_map[ slice_type ];
3567
    if (slice_type == I_TYPE
3568
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
3569
        default_ref_list_done = 1;
3570
    }
3571
    h->slice_type= slice_type;
3572

    
3573
    s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
3574

    
3575
    pps_id= get_ue_golomb(&s->gb);
3576
    if(pps_id>=MAX_PPS_COUNT){
3577
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3578
        return -1;
3579
    }
3580
    if(!h0->pps_buffers[pps_id]) {
3581
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3582
        return -1;
3583
    }
3584
    h->pps= *h0->pps_buffers[pps_id];
3585

    
3586
    if(!h0->sps_buffers[h->pps.sps_id]) {
3587
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3588
        return -1;
3589
    }
3590
    h->sps = *h0->sps_buffers[h->pps.sps_id];
3591

    
3592
    if(h == h0 && h->dequant_coeff_pps != pps_id){
3593
        h->dequant_coeff_pps = pps_id;
3594
        init_dequant_tables(h);
3595
    }
3596

    
3597
    s->mb_width= h->sps.mb_width;
3598
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3599

    
3600
    h->b_stride=  s->mb_width*4;
3601
    h->b8_stride= s->mb_width*2;
3602

    
3603
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3604
    if(h->sps.frame_mbs_only_flag)
3605
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
3606
    else
3607
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
3608

    
3609
    if (s->context_initialized
3610
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3611
        if(h != h0)
3612
            return -1;   // width / height changed during parallelized decoding
3613
        free_tables(h);
3614
        MPV_common_end(s);
3615
    }
3616
    if (!s->context_initialized) {
3617
        if(h != h0)
3618
            return -1;  // we cant (re-)initialize context during parallel decoding
3619
        if (MPV_common_init(s) < 0)
3620
            return -1;
3621

    
3622
        init_scan_tables(h);
3623
        alloc_tables(h);
3624

    
3625
        for(i = 1; i < s->avctx->thread_count; i++) {
3626
            H264Context *c;
3627
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
3628
            memcpy(c, h, sizeof(MpegEncContext));
3629
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
3630
            c->sps = h->sps;
3631
            c->pps = h->pps;
3632
            init_scan_tables(c);
3633
            clone_tables(c, h);
3634
        }
3635

    
3636
        for(i = 0; i < s->avctx->thread_count; i++)
3637
            if(context_init(h->thread_context[i]) < 0)
3638
                return -1;
3639

    
3640
        s->avctx->width = s->width;
3641
        s->avctx->height = s->height;
3642
        s->avctx->sample_aspect_ratio= h->sps.sar;
3643
        if(!s->avctx->sample_aspect_ratio.den)
3644
            s->avctx->sample_aspect_ratio.den = 1;
3645

    
3646
        if(h->sps.timing_info_present_flag){
3647
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale};
3648
            if(h->x264_build > 0 && h->x264_build < 44)
3649
                s->avctx->time_base.den *= 2;
3650
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3651
                      s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
3652
        }
3653
    }
3654

    
3655
    if(h0->current_slice == 0){
3656
        if(frame_start(h) < 0)
3657
            return -1;
3658
    }
3659
    if(h != h0)
3660
        clone_slice(h, h0);
3661

    
3662
    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3663
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3664

    
3665
    h->mb_mbaff = 0;
3666
    h->mb_aff_frame = 0;
3667
    if(h->sps.frame_mbs_only_flag){
3668
        s->picture_structure= PICT_FRAME;
3669
    }else{
3670
        if(get_bits1(&s->gb)) { //field_pic_flag
3671
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3672
            av_log(h->s.avctx, AV_LOG_ERROR, "PAFF interlacing is not implemented\n");
3673
        } else {
3674
            s->picture_structure= PICT_FRAME;
3675
            h->mb_aff_frame = h->sps.mb_aff;
3676
        }
3677
    }
3678
    assert(s->mb_num == s->mb_width * s->mb_height);
3679
    if(first_mb_in_slice << h->mb_aff_frame >= s->mb_num ||
3680
       first_mb_in_slice                    >= s->mb_num){
3681
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3682
        return -1;
3683
    }
3684
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3685
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << h->mb_aff_frame;
3686
    assert(s->mb_y < s->mb_height);
3687

    
3688
    if(s->picture_structure==PICT_FRAME){
3689
        h->curr_pic_num=   h->frame_num;
3690
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3691
    }else{
3692
        h->curr_pic_num= 2*h->frame_num;
3693
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3694
    }
3695

    
3696
    if(h->nal_unit_type == NAL_IDR_SLICE){
3697
        get_ue_golomb(&s->gb); /* idr_pic_id */
3698
    }
3699

    
3700
    if(h->sps.poc_type==0){
3701
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3702

    
3703
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3704
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3705
        }
3706
    }
3707

    
3708
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3709
        h->delta_poc[0]= get_se_golomb(&s->gb);
3710

    
3711
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3712
            h->delta_poc[1]= get_se_golomb(&s->gb);
3713
    }
3714

    
3715
    init_poc(h);
3716

    
3717
    if(h->pps.redundant_pic_cnt_present){
3718
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3719
    }
3720

    
3721
    //set defaults, might be overriden a few line later
3722
    h->ref_count[0]= h->pps.ref_count[0];
3723
    h->ref_count[1]= h->pps.ref_count[1];
3724

    
3725
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3726
        if(h->slice_type == B_TYPE){
3727
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3728
            if(h->sps.mb_aff && h->direct_spatial_mv_pred)
3729
                av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF + spatial direct mode is not implemented\n");
3730
        }
3731
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3732

    
3733
        if(num_ref_idx_active_override_flag){
3734
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3735
            if(h->slice_type==B_TYPE)
3736
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3737

    
3738
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
3739
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3740
                h->ref_count[0]= h->ref_count[1]= 1;
3741
                return -1;
3742
            }
3743
        }
3744
        if(h->slice_type == B_TYPE)
3745
            h->list_count= 2;
3746
        else
3747
            h->list_count= 1;
3748
    }else
3749
        h->list_count= 0;
3750

    
3751
    if(!default_ref_list_done){
3752
        fill_default_ref_list(h);
3753
    }
3754

    
3755
    if(decode_ref_pic_list_reordering(h) < 0)
3756
        return -1;
3757

    
3758
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
3759
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3760
        pred_weight_table(h);
3761
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
3762
        implicit_weight_table(h);
3763
    else
3764
        h->use_weight = 0;
3765

    
3766
    if(s->current_picture.reference)
3767
        decode_ref_pic_marking(h0, &s->gb);
3768

    
3769
    if(FRAME_MBAFF)
3770
        fill_mbaff_ref_list(h);
3771

    
3772
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac ){
3773
        tmp = get_ue_golomb(&s->gb);
3774
        if(tmp > 2){
3775
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3776
            return -1;
3777
        }
3778
        h->cabac_init_idc= tmp;
3779
    }
3780

    
3781
    h->last_qscale_diff = 0;
3782
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3783
    if(tmp>51){
3784
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3785
        return -1;
3786
    }
3787
    s->qscale= tmp;
3788
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3789
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3790
    //FIXME qscale / qp ... stuff
3791
    if(h->slice_type == SP_TYPE){
3792
        get_bits1(&s->gb); /* sp_for_switch_flag */
3793
    }
3794
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3795
        get_se_golomb(&s->gb); /* slice_qs_delta */
3796
    }
3797

    
3798
    h->deblocking_filter = 1;
3799
    h->slice_alpha_c0_offset = 0;
3800
    h->slice_beta_offset = 0;
3801
    if( h->pps.deblocking_filter_parameters_present ) {
3802
        tmp= get_ue_golomb(&s->gb);
3803
        if(tmp > 2){
3804
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3805
            return -1;
3806
        }
3807
        h->deblocking_filter= tmp;
3808
        if(h->deblocking_filter < 2)
3809
            h->deblocking_filter^= 1; // 1<->0
3810

    
3811
        if( h->deblocking_filter ) {
3812
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3813
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3814
        }
3815
    }
3816

    
3817
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
3818
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3819
            /* Cheat slightly for speed:
3820
               Dont bother to deblock across slices */
3821
            h->deblocking_filter = 2;
3822
        } else {
3823
            h0->max_contexts = 1;
3824
            if(!h0->single_decode_warning) {
3825
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3826
                h0->single_decode_warning = 1;
3827
            }
3828
            if(h != h0)
3829
                return 1; // deblocking switched inside frame
3830
        }
3831
    }
3832

    
3833
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
3834
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type != I_TYPE)
3835
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type == B_TYPE)
3836
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3837
        h->deblocking_filter= 0;
3838

    
3839
#if 0 //FMO
3840
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3841
        slice_group_change_cycle= get_bits(&s->gb, ?);
3842
#endif
3843

    
3844
    h0->last_slice_type = slice_type;
3845
    h->slice_num = ++h0->current_slice;
3846

    
3847
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
3848
    h->emu_edge_height= FRAME_MBAFF ? 0 : h->emu_edge_width;
3849

    
3850
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3851
        av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s\n",
3852
               h->slice_num,
3853
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3854
               first_mb_in_slice,
3855
               av_get_pict_type_char(h->slice_type),
3856
               pps_id, h->frame_num,
3857
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3858
               h->ref_count[0], h->ref_count[1],
3859
               s->qscale,
3860
               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
3861
               h->use_weight,
3862
               h->use_weight==1 && h->use_weight_chroma ? "c" : ""
3863
               );
3864
    }
3865

    
3866
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !s->current_picture.reference){
3867
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
3868
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
3869
    }else{
3870
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
3871
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
3872
    }
3873

    
3874
    return 0;
3875
}
3876

    
3877
/**
3878
 *
3879
 */
3880
static inline int get_level_prefix(GetBitContext *gb){
3881
    unsigned int buf;
3882
    int log;
3883

    
3884
    OPEN_READER(re, gb);
3885
    UPDATE_CACHE(re, gb);
3886
    buf=GET_CACHE(re, gb);
3887

    
3888
    log= 32 - av_log2(buf);
3889
#ifdef TRACE
3890
    print_bin(buf>>(32-log), log);
3891
    av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3892
#endif
3893

    
3894
    LAST_SKIP_BITS(re, gb, log);
3895
    CLOSE_READER(re, gb);
3896

    
3897
    return log-1;
3898
}
3899

    
3900
static inline int get_dct8x8_allowed(H264Context *h){
3901
    int i;
3902
    for(i=0; i<4; i++){
3903
        if(!IS_SUB_8X8(h->sub_mb_type[i])
3904
           || (!h->sps.direct_8x8_inference_flag && IS_DIRECT(h->sub_mb_type[i])))
3905
            return 0;
3906
    }
3907
    return 1;
3908
}
3909

    
3910
/**
3911
 * decodes a residual block.
3912
 * @param n block index
3913
 * @param scantable scantable
3914
 * @param max_coeff number of coefficients in the block
3915
 * @return <0 if an error occured
3916
 */
3917
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
3918
    MpegEncContext * const s = &h->s;
3919
    static const int coeff_token_table_index[17]= {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3};
3920
    int level[16];
3921
    int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;
3922

    
3923
    //FIXME put trailing_onex into the context
3924

    
3925
    if(n == CHROMA_DC_BLOCK_INDEX){
3926
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3927
        total_coeff= coeff_token>>2;
3928
    }else{
3929
        if(n == LUMA_DC_BLOCK_INDEX){
3930
            total_coeff= pred_non_zero_count(h, 0);
3931
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3932
            total_coeff= coeff_token>>2;
3933
        }else{
3934
            total_coeff= pred_non_zero_count(h, n);
3935
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3936
            total_coeff= coeff_token>>2;
3937
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3938
        }
3939
    }
3940

    
3941
    //FIXME set last_non_zero?
3942

    
3943
    if(total_coeff==0)
3944
        return 0;
3945
    if(total_coeff > (unsigned)max_coeff) {
3946
        av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
3947
        return -1;
3948
    }
3949

    
3950
    trailing_ones= coeff_token&3;
3951
    tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
3952
    assert(total_coeff<=16);
3953

    
3954
    for(i=0; i<trailing_ones; i++){
3955
        level[i]= 1 - 2*get_bits1(gb);
3956
    }
3957

    
3958
    if(i<total_coeff) {
3959
        int level_code, mask;
3960
        int suffix_length = total_coeff > 10 && trailing_ones < 3;
3961
        int prefix= get_level_prefix(gb);
3962

    
3963
        //first coefficient has suffix_length equal to 0 or 1
3964
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3965
            if(suffix_length)
3966
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3967
            else
3968
                level_code= (prefix<<suffix_length); //part
3969
        }else if(prefix==14){
3970
            if(suffix_length)
3971
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3972
            else
3973
                level_code= prefix + get_bits(gb, 4); //part
3974
        }else if(prefix==15){
3975
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3976
            if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
3977
        }else{
3978
            av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3979
            return -1;
3980
        }
3981

    
3982
        if(trailing_ones < 3) level_code += 2;
3983

    
3984
        suffix_length = 1;
3985
        if(level_code > 5)
3986
            suffix_length++;
3987
        mask= -(level_code&1);
3988
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
3989
        i++;
3990

    
3991
        //remaining coefficients have suffix_length > 0
3992
        for(;i<total_coeff;i++) {
3993
            static const int suffix_limit[7] = {0,5,11,23,47,95,INT_MAX };
3994
            prefix = get_level_prefix(gb);
3995
            if(prefix<15){
3996
                level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
3997
            }else if(prefix==15){
3998
                level_code =  (prefix<<suffix_length) + get_bits(gb, 12);
3999
            }else{
4000
                av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
4001
                return -1;
4002
            }
4003
            mask= -(level_code&1);
4004
            level[i]= (((2+level_code)>>1) ^ mask) - mask;
4005
            if(level_code > suffix_limit[suffix_length])
4006
                suffix_length++;
4007
        }
4008
    }
4009

    
4010
    if(total_coeff == max_coeff)
4011
        zeros_left=0;
4012
    else{
4013
        if(n == CHROMA_DC_BLOCK_INDEX)
4014
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4015
        else
4016
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4017
    }
4018

    
4019
    coeff_num = zeros_left + total_coeff - 1;
4020
    j = scantable[coeff_num];
4021
    if(n > 24){
4022
        block[j] = level[0];
4023
        for(i=1;i<total_coeff;i++) {
4024
            if(zeros_left <= 0)
4025
                run_before = 0;
4026
            else if(zeros_left < 7){
4027
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4028
            }else{
4029
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4030
            }
4031
            zeros_left -= run_before;
4032
            coeff_num -= 1 + run_before;
4033
            j= scantable[ coeff_num ];
4034

    
4035
            block[j]= level[i];
4036
        }
4037
    }else{
4038
        block[j] = (level[0] * qmul[j] + 32)>>6;
4039
        for(i=1;i<total_coeff;i++) {
4040
            if(zeros_left <= 0)
4041
                run_before = 0;
4042
            else if(zeros_left < 7){
4043
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4044
            }else{
4045
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4046
            }
4047
            zeros_left -= run_before;
4048
            coeff_num -= 1 + run_before;
4049
            j= scantable[ coeff_num ];
4050

    
4051
            block[j]= (level[i] * qmul[j] + 32)>>6;
4052
        }
4053
    }
4054

    
4055
    if(zeros_left<0){
4056
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4057
        return -1;
4058
    }
4059

    
4060
    return 0;
4061
}
4062

    
4063
static void predict_field_decoding_flag(H264Context *h){
4064
    MpegEncContext * const s = &h->s;
4065
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4066
    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
4067
                ? s->current_picture.mb_type[mb_xy-1]
4068
                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
4069
                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
4070
                : 0;
4071
    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4072
}
4073

    
4074
/**
4075
 * decodes a P_SKIP or B_SKIP macroblock
4076
 */
4077
static void decode_mb_skip(H264Context *h){
4078
    MpegEncContext * const s = &h->s;
4079
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4080
    int mb_type=0;
4081

    
4082
    memset(h->non_zero_count[mb_xy], 0, 16);
4083
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4084

    
4085
    if(MB_FIELD)
4086
        mb_type|= MB_TYPE_INTERLACED;
4087

    
4088
    if( h->slice_type == B_TYPE )
4089
    {
4090
        // just for fill_caches. pred_direct_motion will set the real mb_type
4091
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4092

    
4093
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4094
        pred_direct_motion(h, &mb_type);
4095
        mb_type|= MB_TYPE_SKIP;
4096
    }
4097
    else
4098
    {
4099
        int mx, my;
4100
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4101

    
4102
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4103
        pred_pskip_motion(h, &mx, &my);
4104
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4105
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4106
    }
4107

    
4108
    write_back_motion(h, mb_type);
4109
    s->current_picture.mb_type[mb_xy]= mb_type;
4110
    s->current_picture.qscale_table[mb_xy]= s->qscale;
4111
    h->slice_table[ mb_xy ]= h->slice_num;
4112
    h->prev_mb_skipped= 1;
4113
}
4114

    
4115
/**
4116
 * decodes a macroblock
4117
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4118
 */
4119
static int decode_mb_cavlc(H264Context *h){
4120
    MpegEncContext * const s = &h->s;
4121
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4122
    int partition_count;
4123
    unsigned int mb_type, cbp;
4124
    int dct8x8_allowed= h->pps.transform_8x8_mode;
4125

    
4126
    s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?
4127

    
4128
    tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4129
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4130
                down the code */
4131
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
4132
        if(s->mb_skip_run==-1)
4133
            s->mb_skip_run= get_ue_golomb(&s->gb);
4134

    
4135
        if (s->mb_skip_run--) {
4136
            if(FRAME_MBAFF && (s->mb_y&1) == 0){
4137
                if(s->mb_skip_run==0)
4138
                    h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4139
                else
4140
                    predict_field_decoding_flag(h);
4141
            }
4142
            decode_mb_skip(h);
4143
            return 0;
4144
        }
4145
    }
4146
    if(FRAME_MBAFF){
4147
        if( (s->mb_y&1) == 0 )
4148
            h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4149
    }else
4150
        h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);
4151

    
4152
    h->prev_mb_skipped= 0;
4153

    
4154
    mb_type= get_ue_golomb(&s->gb);
4155
    if(h->slice_type == B_TYPE){
4156
        if(mb_type < 23){
4157
            partition_count= b_mb_type_info[mb_type].partition_count;
4158
            mb_type=         b_mb_type_info[mb_type].type;
4159
        }else{
4160
            mb_type -= 23;
4161
            goto decode_intra_mb;
4162
        }
4163
    }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
4164
        if(mb_type < 5){
4165
            partition_count= p_mb_type_info[mb_type].partition_count;
4166
            mb_type=         p_mb_type_info[mb_type].type;
4167
        }else{
4168
            mb_type -= 5;
4169
            goto decode_intra_mb;
4170
        }
4171
    }else{
4172
       assert(h->slice_type == I_TYPE);
4173
decode_intra_mb:
4174
        if(mb_type > 25){
4175
            av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice too large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
4176
            return -1;
4177
        }
4178
        partition_count=0;
4179
        cbp= i_mb_type_info[mb_type].cbp;
4180
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4181
        mb_type= i_mb_type_info[mb_type].type;
4182
    }
4183

    
4184
    if(MB_FIELD)
4185
        mb_type |= MB_TYPE_INTERLACED;
4186

    
4187
    h->slice_table[ mb_xy ]= h->slice_num;
4188

    
4189
    if(IS_INTRA_PCM(mb_type)){
4190
        unsigned int x, y;
4191

    
4192
        // We assume these blocks are very rare so we do not optimize it.
4193
        align_get_bits(&s->gb);
4194

    
4195
        // The pixels are stored in the same order as levels in h->mb array.
4196
        for(y=0; y<16; y++){
4197
            const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
4198
            for(x=0; x<16; x++){
4199
                tprintf(s->avctx, "LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4200
                h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8);
4201
            }
4202
        }
4203
        for(y=0; y<8; y++){
4204
            const int index= 256 + 4*(y&3) + 32*(y>>2);
4205
            for(x=0; x<8; x++){
4206
                tprintf(s->avctx, "CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4207
                h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4208
            }
4209
        }
4210
        for(y=0; y<8; y++){
4211
            const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
4212
            for(x=0; x<8; x++){
4213
                tprintf(s->avctx, "CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4214
                h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4215
            }
4216
        }
4217

    
4218
        // In deblocking, the quantizer is 0
4219
        s->current_picture.qscale_table[mb_xy]= 0;
4220
        h->chroma_qp[0] = get_chroma_qp(h, 0, 0);
4221
        h->chroma_qp[1] = get_chroma_qp(h, 1, 0);
4222
        // All coeffs are present
4223
        memset(h->non_zero_count[mb_xy], 16, 16);
4224

    
4225
        s->current_picture.mb_type[mb_xy]= mb_type;
4226
        return 0;
4227
    }
4228

    
4229
    if(MB_MBAFF){
4230
        h->ref_count[0] <<= 1;
4231
        h->ref_count[1] <<= 1;
4232
    }
4233

    
4234
    fill_caches(h, mb_type, 0);
4235

    
4236
    //mb_pred
4237
    if(IS_INTRA(mb_type)){
4238
            int pred_mode;
4239
//            init_top_left_availability(h);
4240
            if(IS_INTRA4x4(mb_type)){
4241
                int i;
4242
                int di = 1;
4243
                if(dct8x8_allowed && get_bits1(&s->gb)){
4244
                    mb_type |= MB_TYPE_8x8DCT;
4245
                    di = 4;
4246
                }
4247

    
4248
//                fill_intra4x4_pred_table(h);
4249
                for(i=0; i<16; i+=di){
4250
                    int mode= pred_intra_mode(h, i);
4251

    
4252
                    if(!get_bits1(&s->gb)){
4253
                        const int rem_mode= get_bits(&s->gb, 3);
4254
                        mode = rem_mode + (rem_mode >= mode);
4255
                    }
4256

    
4257
                    if(di==4)
4258
                        fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
4259
                    else
4260
                        h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4261
                }
4262
                write_back_intra_pred_mode(h);
4263
                if( check_intra4x4_pred_mode(h) < 0)
4264
                    return -1;
4265
            }else{
4266
                h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4267
                if(h->intra16x16_pred_mode < 0)
4268
                    return -1;
4269
            }
4270

    
4271
            pred_mode= check_intra_pred_mode(h, get_ue_golomb(&s->gb));
4272
            if(pred_mode < 0)
4273
                return -1;
4274
            h->chroma_pred_mode= pred_mode;
4275
    }else if(partition_count==4){
4276
        int i, j, sub_partition_count[4], list, ref[2][4];
4277

    
4278
        if(h->slice_type == B_TYPE){
4279
            for(i=0; i<4; i++){
4280
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4281
                if(h->sub_mb_type[i] >=13){
4282
                    av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
4283
                    return -1;
4284
                }
4285
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4286
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4287
            }
4288
            if(   IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4289
               || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
4290
                pred_direct_motion(h, &mb_type);
4291
                h->ref_cache[0][scan8[4]] =
4292
                h->ref_cache[1][scan8[4]] =
4293
                h->ref_cache[0][scan8[12]] =
4294
                h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
4295
            }
4296
        }else{
4297
            assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
4298
            for(i=0; i<4; i++){
4299
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4300
                if(h->sub_mb_type[i] >=4){
4301
                    av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
4302
                    return -1;
4303
                }
4304
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4305
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4306
            }
4307
        }
4308

    
4309
        for(list=0; list<h->list_count; list++){
4310
            int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4311
            for(i=0; i<4; i++){
4312
                if(IS_DIRECT(h->sub_mb_type[i])) continue;
4313
                if(IS_DIR(h->sub_mb_type[i], 0, list)){
4314
                    unsigned int tmp = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4315
                    if(tmp>=ref_count){
4316
                        av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", tmp);
4317
                        return -1;
4318
                    }
4319
                    ref[list][i]= tmp;
4320
                }else{
4321
                 //FIXME
4322
                    ref[list][i] = -1;
4323
                }
4324
            }
4325
        }
4326

    
4327
        if(dct8x8_allowed)
4328
            dct8x8_allowed = get_dct8x8_allowed(h);
4329

    
4330
        for(list=0; list<h->list_count; list++){
4331
            for(i=0; i<4; i++){
4332
                if(IS_DIRECT(h->sub_mb_type[i])) {
4333
                    h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ];
4334
                    continue;
4335
                }
4336
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4337
                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4338

    
4339
                if(IS_DIR(h->sub_mb_type[i], 0, list)){
4340
                    const int sub_mb_type= h->sub_mb_type[i];
4341
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4342
                    for(j=0; j<sub_partition_count[i]; j++){
4343
                        int mx, my;
4344
                        const int index= 4*i + block_width*j;
4345
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4346
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4347
                        mx += get_se_golomb(&s->gb);
4348
                        my += get_se_golomb(&s->gb);
4349
                        tprintf(s->avctx, "final mv:%d %d\n", mx, my);
4350

    
4351
                        if(IS_SUB_8X8(sub_mb_type)){
4352
                            mv_cache[ 1 ][0]=
4353
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4354
                            mv_cache[ 1 ][1]=
4355
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4356
                        }else if(IS_SUB_8X4(sub_mb_type)){
4357
                            mv_cache[ 1 ][0]= mx;
4358
                            mv_cache[ 1 ][1]= my;
4359
                        }else if(IS_SUB_4X8(sub_mb_type)){
4360
                            mv_cache[ 8 ][0]= mx;
4361
                            mv_cache[ 8 ][1]= my;
4362
                        }
4363
                        mv_cache[ 0 ][0]= mx;
4364
                        mv_cache[ 0 ][1]= my;
4365
                    }
4366
                }else{
4367
                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4368
                    p[0] = p[1]=
4369
                    p[8] = p[9]= 0;
4370
                }
4371
            }
4372
        }
4373
    }else if(IS_DIRECT(mb_type)){
4374
        pred_direct_motion(h, &mb_type);
4375
        dct8x8_allowed &= h->sps.direct_8x8_inference_flag;
4376
    }else{
4377
        int list, mx, my, i;
4378
         //FIXME we should set ref_idx_l? to 0 if we use that later ...
4379
        if(IS_16X16(mb_type)){
4380
            for(list=0; list<h->list_count; list++){
4381
                    unsigned int val;
4382
                    if(IS_DIR(mb_type, 0, list)){
4383
                        val= get_te0_golomb(&s->gb, h->ref_count[list]);
4384
                        if(val >= h->ref_count[list]){
4385
                            av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
4386
                            return -1;
4387
                        }
4388
                    }else
4389
                        val= LIST_NOT_USED&0xFF;
4390
                    fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
4391
            }
4392
            for(list=0; list<h->list_count; list++){
4393
                unsigned int val;
4394
                if(IS_DIR(mb_type, 0, list)){
4395
                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
4396
                    mx += get_se_golomb(&s->gb);
4397
                    my += get_se_golomb(&s->gb);
4398
                    tprintf(s->avctx, "final mv:%d %d\n", mx, my);
4399

    
4400
                    val= pack16to32(mx,my);
4401
                }else
4402
                    val=0;
4403
                fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, val, 4);
4404
            }
4405
        }
4406
        else if(IS_16X8(mb_type)){
4407
            for(list=0; list<h->list_count; list++){
4408
                    for(i=0; i<2; i++){
4409
                        unsigned int val;
4410
                        if(IS_DIR(mb_type, i, list)){
4411
                            val= get_te0_golomb(&s->gb, h->ref_count[list]);
4412
                            if(val >= h->ref_count[list]){
4413
                                av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
4414
                                return -1;
4415
                            }
4416
                        }else
4417
                            val= LIST_NOT_USED&0xFF;
4418
                        fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
4419
                    }
4420
            }
4421
            for(list=0; list<h->list_count; list++){
4422
                for(i=0; i<2; i++){
4423
                    unsigned int val;
4424
                    if(IS_DIR(mb_type, i, list)){
4425
                        pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
4426
                        mx += get_se_golomb(&s->gb);
4427
                        my += get_se_golomb(&s->gb);
4428
                        tprintf(s->avctx, "final mv:%d %d\n", mx, my);
4429

    
4430
                        val= pack16to32(mx,my);
4431