Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 02f7695b

History | View | Annotate | Download (311 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
#include "rectangle.h"
36

    
37
#include "cabac.h"
38

    
39
//#undef NDEBUG
40
#include <assert.h>
41

    
42
/**
43
 * Value of Picture.reference when Picture is not a reference picture, but
44
 * is held for delayed output.
45
 */
46
#define DELAYED_PIC_REF 4
47

    
48
static VLC coeff_token_vlc[4];
49
static VLC chroma_dc_coeff_token_vlc;
50

    
51
static VLC total_zeros_vlc[15];
52
static VLC chroma_dc_total_zeros_vlc[3];
53

    
54
static VLC run_vlc[6];
55
static VLC run7_vlc;
56

    
57
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
58
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
59
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);
60
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);
61

    
62
static av_always_inline uint32_t pack16to32(int a, int b){
63
#ifdef WORDS_BIGENDIAN
64
   return (b&0xFFFF) + (a<<16);
65
#else
66
   return (a&0xFFFF) + (b<<16);
67
#endif
68
}
69

    
70
const uint8_t ff_rem6[52]={
71
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,
72
};
73

    
74
const uint8_t ff_div6[52]={
75
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,
76
};
77

    
78

    
79
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
80
    MpegEncContext * const s = &h->s;
81
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
82
    int topleft_xy, top_xy, topright_xy, left_xy[2];
83
    int topleft_type, top_type, topright_type, left_type[2];
84
    int left_block[8];
85
    int topleft_partition= -1;
86
    int i;
87

    
88
    top_xy     = mb_xy  - (s->mb_stride << FIELD_PICTURE);
89

    
90
    //FIXME deblocking could skip the intra and nnz parts.
91
    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
92
        return;
93

    
94
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
95

    
96
    topleft_xy = top_xy - 1;
97
    topright_xy= top_xy + 1;
98
    left_xy[1] = left_xy[0] = mb_xy-1;
99
    left_block[0]= 0;
100
    left_block[1]= 1;
101
    left_block[2]= 2;
102
    left_block[3]= 3;
103
    left_block[4]= 7;
104
    left_block[5]= 10;
105
    left_block[6]= 8;
106
    left_block[7]= 11;
107
    if(FRAME_MBAFF){
108
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
109
        const int top_pair_xy      = pair_xy     - s->mb_stride;
110
        const int topleft_pair_xy  = top_pair_xy - 1;
111
        const int topright_pair_xy = top_pair_xy + 1;
112
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
113
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
114
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
115
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
116
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
117
        const int bottom = (s->mb_y & 1);
118
        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);
119
        if (bottom
120
                ? !curr_mb_frame_flag // bottom macroblock
121
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
122
                ) {
123
            top_xy -= s->mb_stride;
124
        }
125
        if (bottom
126
                ? !curr_mb_frame_flag // bottom macroblock
127
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
128
                ) {
129
            topleft_xy -= s->mb_stride;
130
        } else if(bottom && curr_mb_frame_flag && !left_mb_frame_flag) {
131
            topleft_xy += s->mb_stride;
132
            // take topleft mv from the middle of the mb, as opposed to all other modes which use the bottom-right partition
133
            topleft_partition = 0;
134
        }
135
        if (bottom
136
                ? !curr_mb_frame_flag // bottom macroblock
137
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
138
                ) {
139
            topright_xy -= s->mb_stride;
140
        }
141
        if (left_mb_frame_flag != curr_mb_frame_flag) {
142
            left_xy[1] = left_xy[0] = pair_xy - 1;
143
            if (curr_mb_frame_flag) {
144
                if (bottom) {
145
                    left_block[0]= 2;
146
                    left_block[1]= 2;
147
                    left_block[2]= 3;
148
                    left_block[3]= 3;
149
                    left_block[4]= 8;
150
                    left_block[5]= 11;
151
                    left_block[6]= 8;
152
                    left_block[7]= 11;
153
                } else {
154
                    left_block[0]= 0;
155
                    left_block[1]= 0;
156
                    left_block[2]= 1;
157
                    left_block[3]= 1;
158
                    left_block[4]= 7;
159
                    left_block[5]= 10;
160
                    left_block[6]= 7;
161
                    left_block[7]= 10;
162
                }
163
            } else {
164
                left_xy[1] += s->mb_stride;
165
                //left_block[0]= 0;
166
                left_block[1]= 2;
167
                left_block[2]= 0;
168
                left_block[3]= 2;
169
                //left_block[4]= 7;
170
                left_block[5]= 10;
171
                left_block[6]= 7;
172
                left_block[7]= 10;
173
            }
174
        }
175
    }
176

    
177
    h->top_mb_xy = top_xy;
178
    h->left_mb_xy[0] = left_xy[0];
179
    h->left_mb_xy[1] = left_xy[1];
180
    if(for_deblock){
181
        topleft_type = 0;
182
        topright_type = 0;
183
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
184
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
185
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
186

    
187
        if(FRAME_MBAFF && !IS_INTRA(mb_type)){
188
            int list;
189
            int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
190
            for(i=0; i<16; i++)
191
                h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
192
            for(list=0; list<h->list_count; list++){
193
                if(USES_LIST(mb_type,list)){
194
                    uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
195
                    uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
196
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
197
                    for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
198
                        dst[0] = src[0];
199
                        dst[1] = src[1];
200
                        dst[2] = src[2];
201
                        dst[3] = src[3];
202
                    }
203
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
204
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
205
                    ref += h->b8_stride;
206
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
207
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
208
                }else{
209
                    fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
210
                    fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
211
                }
212
            }
213
        }
214
    }else{
215
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
216
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
217
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
218
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
219
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
220
    }
221

    
222
    if(IS_INTRA(mb_type)){
223
        h->topleft_samples_available=
224
        h->top_samples_available=
225
        h->left_samples_available= 0xFFFF;
226
        h->topright_samples_available= 0xEEEA;
227

    
228
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
229
            h->topleft_samples_available= 0xB3FF;
230
            h->top_samples_available= 0x33FF;
231
            h->topright_samples_available= 0x26EA;
232
        }
233
        for(i=0; i<2; i++){
234
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
235
                h->topleft_samples_available&= 0xDF5F;
236
                h->left_samples_available&= 0x5F5F;
237
            }
238
        }
239

    
240
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
241
            h->topleft_samples_available&= 0x7FFF;
242

    
243
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
244
            h->topright_samples_available&= 0xFBFF;
245

    
246
        if(IS_INTRA4x4(mb_type)){
247
            if(IS_INTRA4x4(top_type)){
248
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
249
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
250
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
251
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
252
            }else{
253
                int pred;
254
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
255
                    pred= -1;
256
                else{
257
                    pred= 2;
258
                }
259
                h->intra4x4_pred_mode_cache[4+8*0]=
260
                h->intra4x4_pred_mode_cache[5+8*0]=
261
                h->intra4x4_pred_mode_cache[6+8*0]=
262
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
263
            }
264
            for(i=0; i<2; i++){
265
                if(IS_INTRA4x4(left_type[i])){
266
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
267
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
268
                }else{
269
                    int pred;
270
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
271
                        pred= -1;
272
                    else{
273
                        pred= 2;
274
                    }
275
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
276
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
277
                }
278
            }
279
        }
280
    }
281

    
282

    
283
/*
284
0 . T T. T T T T
285
1 L . .L . . . .
286
2 L . .L . . . .
287
3 . T TL . . . .
288
4 L . .L . . . .
289
5 L . .. . . . .
290
*/
291
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
292
    if(top_type){
293
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
294
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
295
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
296
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
297

    
298
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
299
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
300

    
301
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
302
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
303

    
304
    }else{
305
        h->non_zero_count_cache[4+8*0]=
306
        h->non_zero_count_cache[5+8*0]=
307
        h->non_zero_count_cache[6+8*0]=
308
        h->non_zero_count_cache[7+8*0]=
309

    
310
        h->non_zero_count_cache[1+8*0]=
311
        h->non_zero_count_cache[2+8*0]=
312

    
313
        h->non_zero_count_cache[1+8*3]=
314
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
315

    
316
    }
317

    
318
    for (i=0; i<2; i++) {
319
        if(left_type[i]){
320
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
321
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
322
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
323
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
324
        }else{
325
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
326
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
327
            h->non_zero_count_cache[0+8*1 +   8*i]=
328
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
329
        }
330
    }
331

    
332
    if( h->pps.cabac ) {
333
        // top_cbp
334
        if(top_type) {
335
            h->top_cbp = h->cbp_table[top_xy];
336
        } else if(IS_INTRA(mb_type)) {
337
            h->top_cbp = 0x1C0;
338
        } else {
339
            h->top_cbp = 0;
340
        }
341
        // left_cbp
342
        if (left_type[0]) {
343
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
344
        } else if(IS_INTRA(mb_type)) {
345
            h->left_cbp = 0x1C0;
346
        } else {
347
            h->left_cbp = 0;
348
        }
349
        if (left_type[0]) {
350
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
351
        }
352
        if (left_type[1]) {
353
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
354
        }
355
    }
356

    
357
#if 1
358
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
359
        int list;
360
        for(list=0; list<h->list_count; list++){
361
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
362
                /*if(!h->mv_cache_clean[list]){
363
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
364
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
365
                    h->mv_cache_clean[list]= 1;
366
                }*/
367
                continue;
368
            }
369
            h->mv_cache_clean[list]= 0;
370

    
371
            if(USES_LIST(top_type, list)){
372
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
373
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
374
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
375
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
376
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
377
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
378
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
379
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
380
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
381
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
382
            }else{
383
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
384
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
385
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
386
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
387
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
388
            }
389

    
390
            for(i=0; i<2; i++){
391
                int cache_idx = scan8[0] - 1 + i*2*8;
392
                if(USES_LIST(left_type[i], list)){
393
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
394
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
395
                    *(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]];
396
                    *(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]];
397
                    h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
398
                    h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
399
                }else{
400
                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
401
                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
402
                    h->ref_cache[list][cache_idx  ]=
403
                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
404
                }
405
            }
406

    
407
            if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
408
                continue;
409

    
410
            if(USES_LIST(topleft_type, list)){
411
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
412
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
413
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
414
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
415
            }else{
416
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
417
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
418
            }
419

    
420
            if(USES_LIST(topright_type, list)){
421
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
422
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
423
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
424
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
425
            }else{
426
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
427
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
428
            }
429

    
430
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
431
                continue;
432

    
433
            h->ref_cache[list][scan8[5 ]+1] =
434
            h->ref_cache[list][scan8[7 ]+1] =
435
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
436
            h->ref_cache[list][scan8[4 ]] =
437
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
438
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
439
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
440
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
441
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
442
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
443

    
444
            if( h->pps.cabac ) {
445
                /* XXX beurk, Load mvd */
446
                if(USES_LIST(top_type, list)){
447
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
448
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
449
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
450
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
451
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
452
                }else{
453
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
454
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
455
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
456
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
457
                }
458
                if(USES_LIST(left_type[0], list)){
459
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
460
                    *(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]];
461
                    *(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]];
462
                }else{
463
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
464
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
465
                }
466
                if(USES_LIST(left_type[1], list)){
467
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
468
                    *(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]];
469
                    *(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]];
470
                }else{
471
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
472
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
473
                }
474
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
475
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
476
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
477
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
478
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
479

    
480
                if(h->slice_type == B_TYPE){
481
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
482

    
483
                    if(IS_DIRECT(top_type)){
484
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
485
                    }else if(IS_8X8(top_type)){
486
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
487
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
488
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
489
                    }else{
490
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
491
                    }
492

    
493
                    if(IS_DIRECT(left_type[0]))
494
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
495
                    else if(IS_8X8(left_type[0]))
496
                        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)];
497
                    else
498
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
499

    
500
                    if(IS_DIRECT(left_type[1]))
501
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
502
                    else if(IS_8X8(left_type[1]))
503
                        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)];
504
                    else
505
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
506
                }
507
            }
508

    
509
            if(FRAME_MBAFF){
510
#define MAP_MVS\
511
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
512
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
513
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
514
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
515
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
516
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
517
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
518
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
519
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
520
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
521
                if(MB_FIELD){
522
#define MAP_F2F(idx, mb_type)\
523
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
524
                        h->ref_cache[list][idx] <<= 1;\
525
                        h->mv_cache[list][idx][1] /= 2;\
526
                        h->mvd_cache[list][idx][1] /= 2;\
527
                    }
528
                    MAP_MVS
529
#undef MAP_F2F
530
                }else{
531
#define MAP_F2F(idx, mb_type)\
532
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
533
                        h->ref_cache[list][idx] >>= 1;\
534
                        h->mv_cache[list][idx][1] <<= 1;\
535
                        h->mvd_cache[list][idx][1] <<= 1;\
536
                    }
537
                    MAP_MVS
538
#undef MAP_F2F
539
                }
540
            }
541
        }
542
    }
543
#endif
544

    
545
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
546
}
547

    
548
static inline void write_back_intra_pred_mode(H264Context *h){
549
    MpegEncContext * const s = &h->s;
550
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
551

    
552
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
553
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
554
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
555
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
556
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
557
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
558
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
559
}
560

    
561
/**
562
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
563
 */
564
static inline int check_intra4x4_pred_mode(H264Context *h){
565
    MpegEncContext * const s = &h->s;
566
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
567
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
568
    int i;
569

    
570
    if(!(h->top_samples_available&0x8000)){
571
        for(i=0; i<4; i++){
572
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
573
            if(status<0){
574
                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);
575
                return -1;
576
            } else if(status){
577
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
578
            }
579
        }
580
    }
581

    
582
    if(!(h->left_samples_available&0x8000)){
583
        for(i=0; i<4; i++){
584
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
585
            if(status<0){
586
                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);
587
                return -1;
588
            } else if(status){
589
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
590
            }
591
        }
592
    }
593

    
594
    return 0;
595
} //FIXME cleanup like next
596

    
597
/**
598
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
599
 */
600
static inline int check_intra_pred_mode(H264Context *h, int mode){
601
    MpegEncContext * const s = &h->s;
602
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
603
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
604

    
605
    if(mode > 6U) {
606
        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);
607
        return -1;
608
    }
609

    
610
    if(!(h->top_samples_available&0x8000)){
611
        mode= top[ mode ];
612
        if(mode<0){
613
            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);
614
            return -1;
615
        }
616
    }
617

    
618
    if(!(h->left_samples_available&0x8000)){
619
        mode= left[ mode ];
620
        if(mode<0){
621
            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);
622
            return -1;
623
        }
624
    }
625

    
626
    return mode;
627
}
628

    
629
/**
630
 * gets the predicted intra4x4 prediction mode.
631
 */
632
static inline int pred_intra_mode(H264Context *h, int n){
633
    const int index8= scan8[n];
634
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
635
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
636
    const int min= FFMIN(left, top);
637

    
638
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
639

    
640
    if(min<0) return DC_PRED;
641
    else      return min;
642
}
643

    
644
static inline void write_back_non_zero_count(H264Context *h){
645
    MpegEncContext * const s = &h->s;
646
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
647

    
648
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
649
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
650
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
651
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
652
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
653
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
654
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
655

    
656
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
657
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
658
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
659

    
660
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
661
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
662
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
663

    
664
    if(FRAME_MBAFF){
665
        // store all luma nnzs, for deblocking
666
        int v = 0, i;
667
        for(i=0; i<16; i++)
668
            v += (!!h->non_zero_count_cache[scan8[i]]) << i;
669
        *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
670
    }
671
}
672

    
673
/**
674
 * gets the predicted number of non zero coefficients.
675
 * @param n block index
676
 */
677
static inline int pred_non_zero_count(H264Context *h, int n){
678
    const int index8= scan8[n];
679
    const int left= h->non_zero_count_cache[index8 - 1];
680
    const int top = h->non_zero_count_cache[index8 - 8];
681
    int i= left + top;
682

    
683
    if(i<64) i= (i+1)>>1;
684

    
685
    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
686

    
687
    return i&31;
688
}
689

    
690
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
691
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
692
    MpegEncContext *s = &h->s;
693

    
694
    /* there is no consistent mapping of mvs to neighboring locations that will
695
     * make mbaff happy, so we can't move all this logic to fill_caches */
696
    if(FRAME_MBAFF){
697
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
698
        const int16_t *mv;
699
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
700
        *C = h->mv_cache[list][scan8[0]-2];
701

    
702
        if(!MB_FIELD
703
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
704
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
705
            if(IS_INTERLACED(mb_types[topright_xy])){
706
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
707
                const int x4 = X4, y4 = Y4;\
708
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
709
                if(!USES_LIST(mb_type,list))\
710
                    return LIST_NOT_USED;\
711
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
712
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
713
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
714
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
715

    
716
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
717
            }
718
        }
719
        if(topright_ref == PART_NOT_AVAILABLE
720
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
721
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
722
            if(!MB_FIELD
723
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
724
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
725
            }
726
            if(MB_FIELD
727
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
728
               && i >= scan8[0]+8){
729
                // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
730
                SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
731
            }
732
        }
733
#undef SET_DIAG_MV
734
    }
735

    
736
    if(topright_ref != PART_NOT_AVAILABLE){
737
        *C= h->mv_cache[list][ i - 8 + part_width ];
738
        return topright_ref;
739
    }else{
740
        tprintf(s->avctx, "topright MV not available\n");
741

    
742
        *C= h->mv_cache[list][ i - 8 - 1 ];
743
        return h->ref_cache[list][ i - 8 - 1 ];
744
    }
745
}
746

    
747
/**
748
 * gets the predicted MV.
749
 * @param n the block index
750
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
751
 * @param mx the x component of the predicted motion vector
752
 * @param my the y component of the predicted motion vector
753
 */
754
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
755
    const int index8= scan8[n];
756
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
757
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
758
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
759
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
760
    const int16_t * C;
761
    int diagonal_ref, match_count;
762

    
763
    assert(part_width==1 || part_width==2 || part_width==4);
764

    
765
/* mv_cache
766
  B . . A T T T T
767
  U . . L . . , .
768
  U . . L . . . .
769
  U . . L . . , .
770
  . . . L . . . .
771
*/
772

    
773
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
774
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
775
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
776
    if(match_count > 1){ //most common
777
        *mx= mid_pred(A[0], B[0], C[0]);
778
        *my= mid_pred(A[1], B[1], C[1]);
779
    }else if(match_count==1){
780
        if(left_ref==ref){
781
            *mx= A[0];
782
            *my= A[1];
783
        }else if(top_ref==ref){
784
            *mx= B[0];
785
            *my= B[1];
786
        }else{
787
            *mx= C[0];
788
            *my= C[1];
789
        }
790
    }else{
791
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
792
            *mx= A[0];
793
            *my= A[1];
794
        }else{
795
            *mx= mid_pred(A[0], B[0], C[0]);
796
            *my= mid_pred(A[1], B[1], C[1]);
797
        }
798
    }
799

    
800
    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);
801
}
802

    
803
/**
804
 * gets the directionally predicted 16x8 MV.
805
 * @param n the block index
806
 * @param mx the x component of the predicted motion vector
807
 * @param my the y component of the predicted motion vector
808
 */
809
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
810
    if(n==0){
811
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
812
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
813

    
814
        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);
815

    
816
        if(top_ref == ref){
817
            *mx= B[0];
818
            *my= B[1];
819
            return;
820
        }
821
    }else{
822
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
823
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
824

    
825
        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);
826

    
827
        if(left_ref == ref){
828
            *mx= A[0];
829
            *my= A[1];
830
            return;
831
        }
832
    }
833

    
834
    //RARE
835
    pred_motion(h, n, 4, list, ref, mx, my);
836
}
837

    
838
/**
839
 * gets the directionally predicted 8x16 MV.
840
 * @param n the block index
841
 * @param mx the x component of the predicted motion vector
842
 * @param my the y component of the predicted motion vector
843
 */
844
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
845
    if(n==0){
846
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
847
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
848

    
849
        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);
850

    
851
        if(left_ref == ref){
852
            *mx= A[0];
853
            *my= A[1];
854
            return;
855
        }
856
    }else{
857
        const int16_t * C;
858
        int diagonal_ref;
859

    
860
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
861

    
862
        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);
863

    
864
        if(diagonal_ref == ref){
865
            *mx= C[0];
866
            *my= C[1];
867
            return;
868
        }
869
    }
870

    
871
    //RARE
872
    pred_motion(h, n, 2, list, ref, mx, my);
873
}
874

    
875
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
876
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
877
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
878

    
879
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
880

    
881
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
882
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
883
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
884

    
885
        *mx = *my = 0;
886
        return;
887
    }
888

    
889
    pred_motion(h, 0, 4, 0, 0, mx, my);
890

    
891
    return;
892
}
893

    
894
static inline void direct_dist_scale_factor(H264Context * const h){
895
    const int poc = h->s.current_picture_ptr->poc;
896
    const int poc1 = h->ref_list[1][0].poc;
897
    int i;
898
    for(i=0; i<h->ref_count[0]; i++){
899
        int poc0 = h->ref_list[0][i].poc;
900
        int td = av_clip(poc1 - poc0, -128, 127);
901
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
902
            h->dist_scale_factor[i] = 256;
903
        }else{
904
            int tb = av_clip(poc - poc0, -128, 127);
905
            int tx = (16384 + (FFABS(td) >> 1)) / td;
906
            h->dist_scale_factor[i] = av_clip((tb*tx + 32) >> 6, -1024, 1023);
907
        }
908
    }
909
    if(FRAME_MBAFF){
910
        for(i=0; i<h->ref_count[0]; i++){
911
            h->dist_scale_factor_field[2*i] =
912
            h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
913
        }
914
    }
915
}
916
static inline void direct_ref_list_init(H264Context * const h){
917
    MpegEncContext * const s = &h->s;
918
    Picture * const ref1 = &h->ref_list[1][0];
919
    Picture * const cur = s->current_picture_ptr;
920
    int list, i, j;
921
    if(cur->pict_type == I_TYPE)
922
        cur->ref_count[0] = 0;
923
    if(cur->pict_type != B_TYPE)
924
        cur->ref_count[1] = 0;
925
    for(list=0; list<2; list++){
926
        cur->ref_count[list] = h->ref_count[list];
927
        for(j=0; j<h->ref_count[list]; j++)
928
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
929
    }
930
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
931
        return;
932
    for(list=0; list<2; list++){
933
        for(i=0; i<ref1->ref_count[list]; i++){
934
            const int poc = ref1->ref_poc[list][i];
935
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
936
            for(j=0; j<h->ref_count[list]; j++)
937
                if(h->ref_list[list][j].poc == poc){
938
                    h->map_col_to_list0[list][i] = j;
939
                    break;
940
                }
941
        }
942
    }
943
    if(FRAME_MBAFF){
944
        for(list=0; list<2; list++){
945
            for(i=0; i<ref1->ref_count[list]; i++){
946
                j = h->map_col_to_list0[list][i];
947
                h->map_col_to_list0_field[list][2*i] = 2*j;
948
                h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
949
            }
950
        }
951
    }
952
}
953

    
954
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
955
    MpegEncContext * const s = &h->s;
956
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
957
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
958
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
959
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
960
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
961
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
962
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
963
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
964
    const int is_b8x8 = IS_8X8(*mb_type);
965
    unsigned int sub_mb_type;
966
    int i8, i4;
967

    
968
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
969
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
970
        /* FIXME save sub mb types from previous frames (or derive from MVs)
971
         * so we know exactly what block size to use */
972
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
973
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
974
    }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
975
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
976
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
977
    }else{
978
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
979
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
980
    }
981
    if(!is_b8x8)
982
        *mb_type |= MB_TYPE_DIRECT2;
983
    if(MB_FIELD)
984
        *mb_type |= MB_TYPE_INTERLACED;
985

    
986
    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);
987

    
988
    if(h->direct_spatial_mv_pred){
989
        int ref[2];
990
        int mv[2][2];
991
        int list;
992

    
993
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
994

    
995
        /* ref = min(neighbors) */
996
        for(list=0; list<2; list++){
997
            int refa = h->ref_cache[list][scan8[0] - 1];
998
            int refb = h->ref_cache[list][scan8[0] - 8];
999
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1000
            if(refc == -2)
1001
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1002
            ref[list] = refa;
1003
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1004
                ref[list] = refb;
1005
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1006
                ref[list] = refc;
1007
            if(ref[list] < 0)
1008
                ref[list] = -1;
1009
        }
1010

    
1011
        if(ref[0] < 0 && ref[1] < 0){
1012
            ref[0] = ref[1] = 0;
1013
            mv[0][0] = mv[0][1] =
1014
            mv[1][0] = mv[1][1] = 0;
1015
        }else{
1016
            for(list=0; list<2; list++){
1017
                if(ref[list] >= 0)
1018
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1019
                else
1020
                    mv[list][0] = mv[list][1] = 0;
1021
            }
1022
        }
1023

    
1024
        if(ref[1] < 0){
1025
            if(!is_b8x8)
1026
                *mb_type &= ~MB_TYPE_L1;
1027
            sub_mb_type &= ~MB_TYPE_L1;
1028
        }else if(ref[0] < 0){
1029
            if(!is_b8x8)
1030
                *mb_type &= ~MB_TYPE_L0;
1031
            sub_mb_type &= ~MB_TYPE_L0;
1032
        }
1033

    
1034
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1035
            int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1036
            int mb_types_col[2];
1037
            int b8_stride = h->b8_stride;
1038
            int b4_stride = h->b_stride;
1039

    
1040
            *mb_type = (*mb_type & ~MB_TYPE_16x16) | MB_TYPE_8x8;
1041

    
1042
            if(IS_INTERLACED(*mb_type)){
1043
                mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1044
                mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1045
                if(s->mb_y&1){
1046
                    l1ref0 -= 2*b8_stride;
1047
                    l1ref1 -= 2*b8_stride;
1048
                    l1mv0 -= 4*b4_stride;
1049
                    l1mv1 -= 4*b4_stride;
1050
                }
1051
                b8_stride *= 3;
1052
                b4_stride *= 6;
1053
            }else{
1054
                int cur_poc = s->current_picture_ptr->poc;
1055
                int *col_poc = h->ref_list[1]->field_poc;
1056
                int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
1057
                int dy = 2*col_parity - (s->mb_y&1);
1058
                mb_types_col[0] =
1059
                mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy + col_parity*s->mb_stride];
1060
                l1ref0 += dy*b8_stride;
1061
                l1ref1 += dy*b8_stride;
1062
                l1mv0 += 2*dy*b4_stride;
1063
                l1mv1 += 2*dy*b4_stride;
1064
                b8_stride = 0;
1065
            }
1066

    
1067
            for(i8=0; i8<4; i8++){
1068
                int x8 = i8&1;
1069
                int y8 = i8>>1;
1070
                int xy8 = x8+y8*b8_stride;
1071
                int xy4 = 3*x8+y8*b4_stride;
1072
                int a=0, b=0;
1073

    
1074
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1075
                    continue;
1076
                h->sub_mb_type[i8] = sub_mb_type;
1077

    
1078
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1079
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1080
                if(!IS_INTRA(mb_types_col[y8])
1081
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1082
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1083
                    if(ref[0] > 0)
1084
                        a= pack16to32(mv[0][0],mv[0][1]);
1085
                    if(ref[1] > 0)
1086
                        b= pack16to32(mv[1][0],mv[1][1]);
1087
                }else{
1088
                    a= pack16to32(mv[0][0],mv[0][1]);
1089
                    b= pack16to32(mv[1][0],mv[1][1]);
1090
                }
1091
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1092
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1093
            }
1094
        }else if(IS_16X16(*mb_type)){
1095
            int a=0, b=0;
1096

    
1097
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1098
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1099
            if(!IS_INTRA(mb_type_col)
1100
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1101
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1102
                       && (h->x264_build>33 || !h->x264_build)))){
1103
                if(ref[0] > 0)
1104
                    a= pack16to32(mv[0][0],mv[0][1]);
1105
                if(ref[1] > 0)
1106
                    b= pack16to32(mv[1][0],mv[1][1]);
1107
            }else{
1108
                a= pack16to32(mv[0][0],mv[0][1]);
1109
                b= pack16to32(mv[1][0],mv[1][1]);
1110
            }
1111
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1112
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1113
        }else{
1114
            for(i8=0; i8<4; i8++){
1115
                const int x8 = i8&1;
1116
                const int y8 = i8>>1;
1117

    
1118
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1119
                    continue;
1120
                h->sub_mb_type[i8] = sub_mb_type;
1121

    
1122
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1123
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1124
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1125
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1126

    
1127
                /* col_zero_flag */
1128
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0
1129
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1130
                                                  && (h->x264_build>33 || !h->x264_build)))){
1131
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1132
                    if(IS_SUB_8X8(sub_mb_type)){
1133
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1134
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1135
                            if(ref[0] == 0)
1136
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1137
                            if(ref[1] == 0)
1138
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1139
                        }
1140
                    }else
1141
                    for(i4=0; i4<4; i4++){
1142
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1143
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1144
                            if(ref[0] == 0)
1145
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1146
                            if(ref[1] == 0)
1147
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1148
                        }
1149
                    }
1150
                }
1151
            }
1152
        }
1153
    }else{ /* direct temporal mv pred */
1154
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1155
        const int *dist_scale_factor = h->dist_scale_factor;
1156

    
1157
        if(FRAME_MBAFF){
1158
            if(IS_INTERLACED(*mb_type)){
1159
                map_col_to_list0[0] = h->map_col_to_list0_field[0];
1160
                map_col_to_list0[1] = h->map_col_to_list0_field[1];
1161
                dist_scale_factor = h->dist_scale_factor_field;
1162
            }
1163
            if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1164
                /* FIXME assumes direct_8x8_inference == 1 */
1165
                const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1166
                int mb_types_col[2];
1167
                int y_shift;
1168

    
1169
                *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1170
                         | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1171
                         | (*mb_type & MB_TYPE_INTERLACED);
1172
                sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1173

    
1174
                if(IS_INTERLACED(*mb_type)){
1175
                    /* frame to field scaling */
1176
                    mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1177
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1178
                    if(s->mb_y&1){
1179
                        l1ref0 -= 2*h->b8_stride;
1180
                        l1ref1 -= 2*h->b8_stride;
1181
                        l1mv0 -= 4*h->b_stride;
1182
                        l1mv1 -= 4*h->b_stride;
1183
                    }
1184
                    y_shift = 0;
1185

    
1186
                    if(   (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1187
                       && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1188
                       && !is_b8x8)
1189
                        *mb_type |= MB_TYPE_16x8;
1190
                    else
1191
                        *mb_type |= MB_TYPE_8x8;
1192
                }else{
1193
                    /* field to frame scaling */
1194
                    /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
1195
                     * but in MBAFF, top and bottom POC are equal */
1196
                    int dy = (s->mb_y&1) ? 1 : 2;
1197
                    mb_types_col[0] =
1198
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1199
                    l1ref0 += dy*h->b8_stride;
1200
                    l1ref1 += dy*h->b8_stride;
1201
                    l1mv0 += 2*dy*h->b_stride;
1202
                    l1mv1 += 2*dy*h->b_stride;
1203
                    y_shift = 2;
1204

    
1205
                    if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1206
                       && !is_b8x8)
1207
                        *mb_type |= MB_TYPE_16x16;
1208
                    else
1209
                        *mb_type |= MB_TYPE_8x8;
1210
                }
1211

    
1212
                for(i8=0; i8<4; i8++){
1213
                    const int x8 = i8&1;
1214
                    const int y8 = i8>>1;
1215
                    int ref0, scale;
1216
                    const int16_t (*l1mv)[2]= l1mv0;
1217

    
1218
                    if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1219
                        continue;
1220
                    h->sub_mb_type[i8] = sub_mb_type;
1221

    
1222
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1223
                    if(IS_INTRA(mb_types_col[y8])){
1224
                        fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1225
                        fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1226
                        fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1227
                        continue;
1228
                    }
1229

    
1230
                    ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1231
                    if(ref0 >= 0)
1232
                        ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1233
                    else{
1234
                        ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1235
                        l1mv= l1mv1;
1236
                    }
1237
                    scale = dist_scale_factor[ref0];
1238
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1239

    
1240
                    {
1241
                        const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
1242
                        int my_col = (mv_col[1]<<y_shift)/2;
1243
                        int mx = (scale * mv_col[0] + 128) >> 8;
1244
                        int my = (scale * my_col + 128) >> 8;
1245
                        fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1246
                        fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1247
                    }
1248
                }
1249
                return;
1250
            }
1251
        }
1252

    
1253
        /* one-to-one mv scaling */
1254

    
1255
        if(IS_16X16(*mb_type)){
1256
            int ref, mv0, mv1;
1257

    
1258
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1259
            if(IS_INTRA(mb_type_col)){
1260
                ref=mv0=mv1=0;
1261
            }else{
1262
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1263
                                                : map_col_to_list0[1][l1ref1[0]];
1264
                const int scale = dist_scale_factor[ref0];
1265
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1266
                int mv_l0[2];
1267
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1268
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1269
                ref= ref0;
1270
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
1271
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1272
            }
1273
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1274
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1275
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1276
        }else{
1277
            for(i8=0; i8<4; i8++){
1278
                const int x8 = i8&1;
1279
                const int y8 = i8>>1;
1280
                int ref0, scale;
1281
                const int16_t (*l1mv)[2]= l1mv0;
1282

    
1283
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1284
                    continue;
1285
                h->sub_mb_type[i8] = sub_mb_type;
1286
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1287
                if(IS_INTRA(mb_type_col)){
1288
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1289
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1290
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1291
                    continue;
1292
                }
1293

    
1294
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1295
                if(ref0 >= 0)
1296
                    ref0 = map_col_to_list0[0][ref0];
1297
                else{
1298
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1299
                    l1mv= l1mv1;
1300
                }
1301
                scale = dist_scale_factor[ref0];
1302

    
1303
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1304
                if(IS_SUB_8X8(sub_mb_type)){
1305
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1306
                    int mx = (scale * mv_col[0] + 128) >> 8;
1307
                    int my = (scale * mv_col[1] + 128) >> 8;
1308
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1309
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1310
                }else
1311
                for(i4=0; i4<4; i4++){
1312
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1313
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1314
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1315
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1316
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1317
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1318
                }
1319
            }
1320
        }
1321
    }
1322
}
1323

    
1324
static inline void write_back_motion(H264Context *h, int mb_type){
1325
    MpegEncContext * const s = &h->s;
1326
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1327
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1328
    int list;
1329

    
1330
    if(!USES_LIST(mb_type, 0))
1331
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1332

    
1333
    for(list=0; list<h->list_count; list++){
1334
        int y;
1335
        if(!USES_LIST(mb_type, list))
1336
            continue;
1337

    
1338
        for(y=0; y<4; y++){
1339
            *(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];
1340
            *(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];
1341
        }
1342
        if( h->pps.cabac ) {
1343
            if(IS_SKIP(mb_type))
1344
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1345
            else
1346
            for(y=0; y<4; y++){
1347
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1348
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1349
            }
1350
        }
1351

    
1352
        {
1353
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1354
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1355
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1356
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1357
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1358
        }
1359
    }
1360

    
1361
    if(h->slice_type == B_TYPE && h->pps.cabac){
1362
        if(IS_8X8(mb_type)){
1363
            uint8_t *direct_table = &h->direct_table[b8_xy];
1364
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1365
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1366
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1367
        }
1368
    }
1369
}
1370

    
1371
/**
1372
 * Decodes a network abstraction layer unit.
1373
 * @param consumed is the number of bytes used as input
1374
 * @param length is the length of the array
1375
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1376
 * @returns decoded bytes, might be src+1 if no escapes
1377
 */
1378
static const uint8_t *decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1379
    int i, si, di;
1380
    uint8_t *dst;
1381
    int bufidx;
1382

    
1383
//    src[0]&0x80;                //forbidden bit
1384
    h->nal_ref_idc= src[0]>>5;
1385
    h->nal_unit_type= src[0]&0x1F;
1386

    
1387
    src++; length--;
1388
#if 0
1389
    for(i=0; i<length; i++)
1390
        printf("%2X ", src[i]);
1391
#endif
1392
    for(i=0; i+1<length; i+=2){
1393
        if(src[i]) continue;
1394
        if(i>0 && src[i-1]==0) i--;
1395
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1396
            if(src[i+2]!=3){
1397
                /* startcode, so we must be past the end */
1398
                length=i;
1399
            }
1400
            break;
1401
        }
1402
    }
1403

    
1404
    if(i>=length-1){ //no escaped 0
1405
        *dst_length= length;
1406
        *consumed= length+1; //+1 for the header
1407
        return src;
1408
    }
1409

    
1410
    bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1411
    h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length);
1412
    dst= h->rbsp_buffer[bufidx];
1413

    
1414
    if (dst == NULL){
1415
        return NULL;
1416
    }
1417

    
1418
//printf("decoding esc\n");
1419
    si=di=0;
1420
    while(si<length){
1421
        //remove escapes (very rare 1:2^22)
1422
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1423
            if(src[si+2]==3){ //escape
1424
                dst[di++]= 0;
1425
                dst[di++]= 0;
1426
                si+=3;
1427
                continue;
1428
            }else //next start code
1429
                break;
1430
        }
1431

    
1432
        dst[di++]= src[si++];
1433
    }
1434

    
1435
    *dst_length= di;
1436
    *consumed= si + 1;//+1 for the header
1437
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1438
    return dst;
1439
}
1440

    
1441
/**
1442
 * identifies the exact end of the bitstream
1443
 * @return the length of the trailing, or 0 if damaged
1444
 */
1445
static int decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1446
    int v= *src;
1447
    int r;
1448

    
1449
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1450

    
1451
    for(r=1; r<9; r++){
1452
        if(v&1) return r;
1453
        v>>=1;
1454
    }
1455
    return 0;
1456
}
1457

    
1458
/**
1459
 * idct tranforms the 16 dc values and dequantize them.
1460
 * @param qp quantization parameter
1461
 */
1462
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1463
#define stride 16
1464
    int i;
1465
    int temp[16]; //FIXME check if this is a good idea
1466
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1467
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1468

    
1469
//memset(block, 64, 2*256);
1470
//return;
1471
    for(i=0; i<4; i++){
1472
        const int offset= y_offset[i];
1473
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1474
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1475
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1476
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1477

    
1478
        temp[4*i+0]= z0+z3;
1479
        temp[4*i+1]= z1+z2;
1480
        temp[4*i+2]= z1-z2;
1481
        temp[4*i+3]= z0-z3;
1482
    }
1483

    
1484
    for(i=0; i<4; i++){
1485
        const int offset= x_offset[i];
1486
        const int z0= temp[4*0+i] + temp[4*2+i];
1487
        const int z1= temp[4*0+i] - temp[4*2+i];
1488
        const int z2= temp[4*1+i] - temp[4*3+i];
1489
        const int z3= temp[4*1+i] + temp[4*3+i];
1490

    
1491
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1492
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1493
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1494
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1495
    }
1496
}
1497

    
1498
#if 0
1499
/**
1500
 * dct tranforms the 16 dc values.
1501
 * @param qp quantization parameter ??? FIXME
1502
 */
1503
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1504
//    const int qmul= dequant_coeff[qp][0];
1505
    int i;
1506
    int temp[16]; //FIXME check if this is a good idea
1507
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1508
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1509

1510
    for(i=0; i<4; i++){
1511
        const int offset= y_offset[i];
1512
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1513
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1514
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1515
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1516

1517
        temp[4*i+0]= z0+z3;
1518
        temp[4*i+1]= z1+z2;
1519
        temp[4*i+2]= z1-z2;
1520
        temp[4*i+3]= z0-z3;
1521
    }
1522

1523
    for(i=0; i<4; i++){
1524
        const int offset= x_offset[i];
1525
        const int z0= temp[4*0+i] + temp[4*2+i];
1526
        const int z1= temp[4*0+i] - temp[4*2+i];
1527
        const int z2= temp[4*1+i] - temp[4*3+i];
1528
        const int z3= temp[4*1+i] + temp[4*3+i];
1529

1530
        block[stride*0 +offset]= (z0 + z3)>>1;
1531
        block[stride*2 +offset]= (z1 + z2)>>1;
1532
        block[stride*8 +offset]= (z1 - z2)>>1;
1533
        block[stride*10+offset]= (z0 - z3)>>1;
1534
    }
1535
}
1536
#endif
1537

    
1538
#undef xStride
1539
#undef stride
1540

    
1541
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1542
    const int stride= 16*2;
1543
    const int xStride= 16;
1544
    int a,b,c,d,e;
1545

    
1546
    a= block[stride*0 + xStride*0];
1547
    b= block[stride*0 + xStride*1];
1548
    c= block[stride*1 + xStride*0];
1549
    d= block[stride*1 + xStride*1];
1550

    
1551
    e= a-b;
1552
    a= a+b;
1553
    b= c-d;
1554
    c= c+d;
1555

    
1556
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1557
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1558
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1559
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1560
}
1561

    
1562
#if 0
1563
static void chroma_dc_dct_c(DCTELEM *block){
1564
    const int stride= 16*2;
1565
    const int xStride= 16;
1566
    int a,b,c,d,e;
1567

1568
    a= block[stride*0 + xStride*0];
1569
    b= block[stride*0 + xStride*1];
1570
    c= block[stride*1 + xStride*0];
1571
    d= block[stride*1 + xStride*1];
1572

1573
    e= a-b;
1574
    a= a+b;
1575
    b= c-d;
1576
    c= c+d;
1577

1578
    block[stride*0 + xStride*0]= (a+c);
1579
    block[stride*0 + xStride*1]= (e+b);
1580
    block[stride*1 + xStride*0]= (a-c);
1581
    block[stride*1 + xStride*1]= (e-b);
1582
}
1583
#endif
1584

    
1585
/**
1586
 * gets the chroma qp.
1587
 */
1588
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1589
    return h->pps.chroma_qp_table[t][qscale & 0xff];
1590
}
1591

    
1592
//FIXME need to check that this does not overflow signed 32 bit for low qp, i am not sure, it's very close
1593
//FIXME check that gcc inlines this (and optimizes intra & separate_dc stuff away)
1594
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int separate_dc){
1595
    int i;
1596
    const int * const quant_table= quant_coeff[qscale];
1597
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1598
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1599
    const unsigned int threshold2= (threshold1<<1);
1600
    int last_non_zero;
1601

    
1602
    if(separate_dc){
1603
        if(qscale<=18){
1604
            //avoid overflows
1605
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1606
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1607
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1608

    
1609
            int level= block[0]*quant_coeff[qscale+18][0];
1610
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1611
                if(level>0){
1612
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1613
                    block[0]= level;
1614
                }else{
1615
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1616
                    block[0]= -level;
1617
                }
1618
//                last_non_zero = i;
1619
            }else{
1620
                block[0]=0;
1621
            }
1622
        }else{
1623
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1624
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1625
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1626

    
1627
            int level= block[0]*quant_table[0];
1628
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1629
                if(level>0){
1630
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1631
                    block[0]= level;
1632
                }else{
1633
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1634
                    block[0]= -level;
1635
                }
1636
//                last_non_zero = i;
1637
            }else{
1638
                block[0]=0;
1639
            }
1640
        }
1641
        last_non_zero= 0;
1642
        i=1;
1643
    }else{
1644
        last_non_zero= -1;
1645
        i=0;
1646
    }
1647

    
1648
    for(; i<16; i++){
1649
        const int j= scantable[i];
1650
        int level= block[j]*quant_table[j];
1651

    
1652
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1653
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1654
        if(((unsigned)(level+threshold1))>threshold2){
1655
            if(level>0){
1656
                level= (bias + level)>>QUANT_SHIFT;
1657
                block[j]= level;
1658
            }else{
1659
                level= (bias - level)>>QUANT_SHIFT;
1660
                block[j]= -level;
1661
            }
1662
            last_non_zero = i;
1663
        }else{
1664
            block[j]=0;
1665
        }
1666
    }
1667

    
1668
    return last_non_zero;
1669
}
1670

    
1671
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1672
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1673
                           int src_x_offset, int src_y_offset,
1674
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1675
    MpegEncContext * const s = &h->s;
1676
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1677
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1678
    const int luma_xy= (mx&3) + ((my&3)<<2);
1679
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1680
    uint8_t * src_cb, * src_cr;
1681
    int extra_width= h->emu_edge_width;
1682
    int extra_height= h->emu_edge_height;
1683
    int emu=0;
1684
    const int full_mx= mx>>2;
1685
    const int full_my= my>>2;
1686
    const int pic_width  = 16*s->mb_width;
1687
    const int pic_height = 16*s->mb_height >> MB_FIELD;
1688

    
1689
    if(!pic->data[0]) //FIXME this is unacceptable, some senseable error concealment must be done for missing reference frames
1690
        return;
1691

    
1692
    if(mx&7) extra_width -= 3;
1693
    if(my&7) extra_height -= 3;
1694

    
1695
    if(   full_mx < 0-extra_width
1696
       || full_my < 0-extra_height
1697
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
1698
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
1699
        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);
1700
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1701
        emu=1;
1702
    }
1703

    
1704
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1705
    if(!square){
1706
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1707
    }
1708

    
1709
    if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1710

    
1711
    if(MB_FIELD){
1712
        // chroma offset when predicting from a field of opposite parity
1713
        my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1714
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1715
    }
1716
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1717
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1718

    
1719
    if(emu){
1720
        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);
1721
            src_cb= s->edge_emu_buffer;
1722
    }
1723
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1724

    
1725
    if(emu){
1726
        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);
1727
            src_cr= s->edge_emu_buffer;
1728
    }
1729
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1730
}
1731

    
1732
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1733
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1734
                           int x_offset, int y_offset,
1735
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1736
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1737
                           int list0, int list1){
1738
    MpegEncContext * const s = &h->s;
1739
    qpel_mc_func *qpix_op=  qpix_put;
1740
    h264_chroma_mc_func chroma_op= chroma_put;
1741

    
1742
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1743
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1744
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1745
    x_offset += 8*s->mb_x;
1746
    y_offset += 8*(s->mb_y >> MB_FIELD);
1747

    
1748
    if(list0){
1749
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1750
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1751
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1752
                           qpix_op, chroma_op);
1753

    
1754
        qpix_op=  qpix_avg;
1755
        chroma_op= chroma_avg;
1756
    }
1757

    
1758
    if(list1){
1759
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1760
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1761
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1762
                           qpix_op, chroma_op);
1763
    }
1764
}
1765

    
1766
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1767
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1768
                           int x_offset, int y_offset,
1769
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1770
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1771
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1772
                           int list0, int list1){
1773
    MpegEncContext * const s = &h->s;
1774

    
1775
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1776
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1777
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1778
    x_offset += 8*s->mb_x;
1779
    y_offset += 8*(s->mb_y >> MB_FIELD);
1780

    
1781
    if(list0 && list1){
1782
        /* don't optimize for luma-only case, since B-frames usually
1783
         * use implicit weights => chroma too. */
1784
        uint8_t *tmp_cb = s->obmc_scratchpad;
1785
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1786
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1787
        int refn0 = h->ref_cache[0][ scan8[n] ];
1788
        int refn1 = h->ref_cache[1][ scan8[n] ];
1789

    
1790
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1791
                    dest_y, dest_cb, dest_cr,
1792
                    x_offset, y_offset, qpix_put, chroma_put);
1793
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1794
                    tmp_y, tmp_cb, tmp_cr,
1795
                    x_offset, y_offset, qpix_put, chroma_put);
1796

    
1797
        if(h->use_weight == 2){
1798
            int weight0 = h->implicit_weight[refn0][refn1];
1799
            int weight1 = 64 - weight0;
1800
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
1801
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1802
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1803
        }else{
1804
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1805
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1806
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1807
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1808
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1809
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1810
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1811
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1812
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1813
        }
1814
    }else{
1815
        int list = list1 ? 1 : 0;
1816
        int refn = h->ref_cache[list][ scan8[n] ];
1817
        Picture *ref= &h->ref_list[list][refn];
1818
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1819
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
1820
                    qpix_put, chroma_put);
1821

    
1822
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1823
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
1824
        if(h->use_weight_chroma){
1825
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1826
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1827
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1828
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1829
        }
1830
    }
1831
}
1832

    
1833
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1834
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1835
                           int x_offset, int y_offset,
1836
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1837
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1838
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1839
                           int list0, int list1){
1840
    if((h->use_weight==2 && list0 && list1
1841
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1842
       || h->use_weight==1)
1843
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1844
                         x_offset, y_offset, qpix_put, chroma_put,
1845
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1846
    else
1847
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1848
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1849
}
1850

    
1851
static inline void prefetch_motion(H264Context *h, int list){
1852
    /* fetch pixels for estimated mv 4 macroblocks ahead
1853
     * optimized for 64byte cache lines */
1854
    MpegEncContext * const s = &h->s;
1855
    const int refn = h->ref_cache[list][scan8[0]];
1856
    if(refn >= 0){
1857
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1858
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1859
        uint8_t **src= h->ref_list[list][refn].data;
1860
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1861
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
1862
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1863
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1864
    }
1865
}
1866

    
1867
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1868
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1869
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1870
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1871
    MpegEncContext * const s = &h->s;
1872
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1873
    const int mb_type= s->current_picture.mb_type[mb_xy];
1874

    
1875
    assert(IS_INTER(mb_type));
1876

    
1877
    prefetch_motion(h, 0);
1878

    
1879
    if(IS_16X16(mb_type)){
1880
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1881
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1882
                &weight_op[0], &weight_avg[0],
1883
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1884
    }else if(IS_16X8(mb_type)){
1885
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1886
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1887
                &weight_op[1], &weight_avg[1],
1888
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1889
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1890
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1891
                &weight_op[1], &weight_avg[1],
1892
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1893
    }else if(IS_8X16(mb_type)){
1894
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1895
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1896
                &weight_op[2], &weight_avg[2],
1897
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1898
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1899
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1900
                &weight_op[2], &weight_avg[2],
1901
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1902
    }else{
1903
        int i;
1904

    
1905
        assert(IS_8X8(mb_type));
1906

    
1907
        for(i=0; i<4; i++){
1908
            const int sub_mb_type= h->sub_mb_type[i];
1909
            const int n= 4*i;
1910
            int x_offset= (i&1)<<2;
1911
            int y_offset= (i&2)<<1;
1912

    
1913
            if(IS_SUB_8X8(sub_mb_type)){
1914
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1915
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1916
                    &weight_op[3], &weight_avg[3],
1917
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1918
            }else if(IS_SUB_8X4(sub_mb_type)){
1919
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1920
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1921
                    &weight_op[4], &weight_avg[4],
1922
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1923
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1924
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1925
                    &weight_op[4], &weight_avg[4],
1926
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1927
            }else if(IS_SUB_4X8(sub_mb_type)){
1928
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1929
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1930
                    &weight_op[5], &weight_avg[5],
1931
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1932
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1933
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1934
                    &weight_op[5], &weight_avg[5],
1935
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1936
            }else{
1937
                int j;
1938
                assert(IS_SUB_4X4(sub_mb_type));
1939
                for(j=0; j<4; j++){
1940
                    int sub_x_offset= x_offset + 2*(j&1);
1941
                    int sub_y_offset= y_offset +   (j&2);
1942
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1943
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1944
                        &weight_op[6], &weight_avg[6],
1945
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1946
                }
1947
            }
1948
        }
1949
    }
1950

    
1951
    prefetch_motion(h, 1);
1952
}
1953

    
1954
static void decode_init_vlc(void){
1955
    static int done = 0;
1956

    
1957
    if (!done) {
1958
        int i;
1959
        done = 1;
1960

    
1961
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1962
                 &chroma_dc_coeff_token_len [0], 1, 1,
1963
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
1964

    
1965
        for(i=0; i<4; i++){
1966
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1967
                     &coeff_token_len [i][0], 1, 1,
1968
                     &coeff_token_bits[i][0], 1, 1, 1);
1969
        }
1970

    
1971
        for(i=0; i<3; i++){
1972
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1973
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
1974
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
1975
        }
1976
        for(i=0; i<15; i++){
1977
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
1978
                     &total_zeros_len [i][0], 1, 1,
1979
                     &total_zeros_bits[i][0], 1, 1, 1);
1980
        }
1981

    
1982
        for(i=0; i<6; i++){
1983
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
1984
                     &run_len [i][0], 1, 1,
1985
                     &run_bits[i][0], 1, 1, 1);
1986
        }
1987
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1988
                 &run_len [6][0], 1, 1,
1989
                 &run_bits[6][0], 1, 1, 1);
1990
    }
1991
}
1992

    
1993
static void free_tables(H264Context *h){
1994
    int i;
1995
    H264Context *hx;
1996
    av_freep(&h->intra4x4_pred_mode);
1997
    av_freep(&h->chroma_pred_mode_table);
1998
    av_freep(&h->cbp_table);
1999
    av_freep(&h->mvd_table[0]);
2000
    av_freep(&h->mvd_table[1]);
2001
    av_freep(&h->direct_table);
2002
    av_freep(&h->non_zero_count);
2003
    av_freep(&h->slice_table_base);
2004
    h->slice_table= NULL;
2005

    
2006
    av_freep(&h->mb2b_xy);
2007
    av_freep(&h->mb2b8_xy);
2008

    
2009
    for(i = 0; i < MAX_SPS_COUNT; i++)
2010
        av_freep(h->sps_buffers + i);
2011

    
2012
    for(i = 0; i < MAX_PPS_COUNT; i++)
2013
        av_freep(h->pps_buffers + i);
2014

    
2015
    for(i = 0; i < h->s.avctx->thread_count; i++) {
2016
        hx = h->thread_context[i];
2017
        if(!hx) continue;
2018
        av_freep(&hx->top_borders[1]);
2019
        av_freep(&hx->top_borders[0]);
2020
        av_freep(&hx->s.obmc_scratchpad);
2021
    }
2022
}
2023

    
2024
static void init_dequant8_coeff_table(H264Context *h){
2025
    int i,q,x;
2026
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2027
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
2028
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
2029

    
2030
    for(i=0; i<2; i++ ){
2031
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2032
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
2033
            break;
2034
        }
2035

    
2036
        for(q=0; q<52; q++){
2037
            int shift = ff_div6[q];
2038
            int idx = ff_rem6[q];
2039
            for(x=0; x<64; x++)
2040
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2041
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2042
                    h->pps.scaling_matrix8[i][x]) << shift;
2043
        }
2044
    }
2045
}
2046

    
2047
static void init_dequant4_coeff_table(H264Context *h){
2048
    int i,j,q,x;
2049
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2050
    for(i=0; i<6; i++ ){
2051
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
2052
        for(j=0; j<i; j++){
2053
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2054
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2055
                break;
2056
            }
2057
        }
2058
        if(j<i)
2059
            continue;
2060

    
2061
        for(q=0; q<52; q++){
2062
            int shift = ff_div6[q] + 2;
2063
            int idx = ff_rem6[q];
2064
            for(x=0; x<16; x++)
2065
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2066
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2067
                    h->pps.scaling_matrix4[i][x]) << shift;
2068
        }
2069
    }
2070
}
2071

    
2072
static void init_dequant_tables(H264Context *h){
2073
    int i,x;
2074
    init_dequant4_coeff_table(h);
2075
    if(h->pps.transform_8x8_mode)
2076
        init_dequant8_coeff_table(h);
2077
    if(h->sps.transform_bypass){
2078
        for(i=0; i<6; i++)
2079
            for(x=0; x<16; x++)
2080
                h->dequant4_coeff[i][0][x] = 1<<6;
2081
        if(h->pps.transform_8x8_mode)
2082
            for(i=0; i<2; i++)
2083
                for(x=0; x<64; x++)
2084
                    h->dequant8_coeff[i][0][x] = 1<<6;
2085
    }
2086
}
2087

    
2088

    
2089
/**
2090
 * allocates tables.
2091
 * needs width/height
2092
 */
2093
static int alloc_tables(H264Context *h){
2094
    MpegEncContext * const s = &h->s;
2095
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2096
    int x,y;
2097

    
2098
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2099

    
2100
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2101
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
2102
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2103

    
2104
    CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2105
    CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2106
    CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2107
    CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2108

    
2109
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));
2110
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2111

    
2112
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2113
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2114
    for(y=0; y<s->mb_height; y++){
2115
        for(x=0; x<s->mb_width; x++){
2116
            const int mb_xy= x + y*s->mb_stride;
2117
            const int b_xy = 4*x + 4*y*h->b_stride;
2118
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2119

    
2120
            h->mb2b_xy [mb_xy]= b_xy;
2121
            h->mb2b8_xy[mb_xy]= b8_xy;
2122
        }
2123
    }
2124

    
2125
    s->obmc_scratchpad = NULL;
2126

    
2127
    if(!h->dequant4_coeff[0])
2128
        init_dequant_tables(h);
2129

    
2130
    return 0;
2131
fail:
2132
    free_tables(h);
2133
    return -1;
2134
}
2135

    
2136
/**
2137
 * Mimic alloc_tables(), but for every context thread.
2138
 */
2139
static void clone_tables(H264Context *dst, H264Context *src){
2140
    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
2141
    dst->non_zero_count           = src->non_zero_count;
2142
    dst->slice_table              = src->slice_table;
2143
    dst->cbp_table                = src->cbp_table;
2144
    dst->mb2b_xy                  = src->mb2b_xy;
2145
    dst->mb2b8_xy                 = src->mb2b8_xy;
2146
    dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
2147
    dst->mvd_table[0]             = src->mvd_table[0];
2148
    dst->mvd_table[1]             = src->mvd_table[1];
2149
    dst->direct_table             = src->direct_table;
2150

    
2151
    dst->s.obmc_scratchpad = NULL;
2152
    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2153
}
2154

    
2155
/**
2156
 * Init context
2157
 * Allocate buffers which are not shared amongst multiple threads.
2158
 */
2159
static int context_init(H264Context *h){
2160
    CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2161
    CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2162

    
2163
    return 0;
2164
fail:
2165
    return -1; // free_tables will clean up for us
2166
}
2167

    
2168
static void common_init(H264Context *h){
2169
    MpegEncContext * const s = &h->s;
2170

    
2171
    s->width = s->avctx->width;
2172
    s->height = s->avctx->height;
2173
    s->codec_id= s->avctx->codec->id;
2174

    
2175
    ff_h264_pred_init(&h->hpc, s->codec_id);
2176

    
2177
    h->dequant_coeff_pps= -1;
2178
    s->unrestricted_mv=1;
2179
    s->decode=1; //FIXME
2180

    
2181
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2182
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2183
}
2184

    
2185
static int decode_init(AVCodecContext *avctx){
2186
    H264Context *h= avctx->priv_data;
2187
    MpegEncContext * const s = &h->s;
2188

    
2189
    MPV_decode_defaults(s);
2190

    
2191
    s->avctx = avctx;
2192
    common_init(h);
2193

    
2194
    s->out_format = FMT_H264;
2195
    s->workaround_bugs= avctx->workaround_bugs;
2196

    
2197
    // set defaults
2198
//    s->decode_mb= ff_h263_decode_mb;
2199
    s->quarter_sample = 1;
2200
    s->low_delay= 1;
2201
    avctx->pix_fmt= PIX_FMT_YUV420P;
2202

    
2203
    decode_init_vlc();
2204

    
2205
    if(avctx->extradata_size > 0 && avctx->extradata &&
2206
       *(char *)avctx->extradata == 1){
2207
        h->is_avc = 1;
2208
        h->got_avcC = 0;
2209
    } else {
2210
        h->is_avc = 0;
2211
    }
2212

    
2213
    h->thread_context[0] = h;
2214
    return 0;
2215
}
2216

    
2217
static int frame_start(H264Context *h){
2218
    MpegEncContext * const s = &h->s;
2219
    int i;
2220

    
2221
    if(MPV_frame_start(s, s->avctx) < 0)
2222
        return -1;
2223
    ff_er_frame_start(s);
2224
    /*
2225
     * MPV_frame_start uses pict_type to derive key_frame.
2226
     * This is incorrect for H.264; IDR markings must be used.
2227
     * Zero here; IDR markings per slice in frame or fields are OR'd in later.
2228
     * See decode_nal_units().
2229
     */
2230
    s->current_picture_ptr->key_frame= 0;
2231

    
2232
    assert(s->linesize && s->uvlinesize);
2233

    
2234
    for(i=0; i<16; i++){
2235
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2236
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2237
    }
2238
    for(i=0; i<4; i++){
2239
        h->block_offset[16+i]=
2240
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2241
        h->block_offset[24+16+i]=
2242
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2243
    }
2244

    
2245
    /* can't be in alloc_tables because linesize isn't known there.
2246
     * FIXME: redo bipred weight to not require extra buffer? */
2247
    for(i = 0; i < s->avctx->thread_count; i++)
2248
        if(!h->thread_context[i]->s.obmc_scratchpad)
2249
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2250

    
2251
    /* some macroblocks will be accessed before they're available */
2252
    if(FRAME_MBAFF || s->avctx->thread_count > 1)
2253
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
2254

    
2255
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2256
    return 0;
2257
}
2258

    
2259
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){
2260
    MpegEncContext * const s = &h->s;
2261
    int i;
2262

    
2263
    src_y  -=   linesize;
2264
    src_cb -= uvlinesize;
2265
    src_cr -= uvlinesize;
2266

    
2267
    // There are two lines saved, the line above the the top macroblock of a pair,
2268
    // and the line above the bottom macroblock
2269
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
2270
    for(i=1; i<17; i++){
2271
        h->left_border[i]= src_y[15+i*  linesize];
2272
    }
2273

    
2274
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2275
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2276

    
2277
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2278
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
2279
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
2280
        for(i=1; i<9; i++){
2281
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
2282
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2283
        }
2284
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2285
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2286
    }
2287
}
2288

    
2289
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){
2290
    MpegEncContext * const s = &h->s;
2291
    int temp8, i;
2292
    uint64_t temp64;
2293
    int deblock_left;
2294
    int deblock_top;
2295
    int mb_xy;
2296

    
2297
    if(h->deblocking_filter == 2) {
2298
        mb_xy = s->mb_x + s->mb_y*s->mb_stride;
2299
        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2300
        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2301
    } else {
2302
        deblock_left = (s->mb_x > 0);
2303
        deblock_top =  (s->mb_y > 0);
2304
    }
2305

    
2306
    src_y  -=   linesize + 1;
2307
    src_cb -= uvlinesize + 1;
2308
    src_cr -= uvlinesize + 1;
2309

    
2310
#define XCHG(a,b,t,xchg)\
2311
t= a;\
2312
if(xchg)\
2313
    a= b;\
2314
b= t;
2315

    
2316
    if(deblock_left){
2317
        for(i = !deblock_top; i<17; i++){
2318
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2319
        }
2320
    }
2321

    
2322
    if(deblock_top){
2323
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2324
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2325
        if(s->mb_x+1 < s->mb_width){
2326
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2327
        }
2328
    }
2329

    
2330
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2331
        if(deblock_left){
2332
            for(i = !deblock_top; i<9; i++){
2333
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
2334
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2335
            }
2336
        }
2337
        if(deblock_top){
2338
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2339
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2340
        }
2341
    }
2342
}
2343

    
2344
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2345
    MpegEncContext * const s = &h->s;
2346
    int i;
2347

    
2348
    src_y  -= 2 *   linesize;
2349
    src_cb -= 2 * uvlinesize;
2350
    src_cr -= 2 * uvlinesize;
2351

    
2352
    // There are two lines saved, the line above the the top macroblock of a pair,
2353
    // and the line above the bottom macroblock
2354
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
2355
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
2356
    for(i=2; i<34; i++){
2357
        h->left_border[i]= src_y[15+i*  linesize];
2358
    }
2359

    
2360
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
2361
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
2362
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
2363
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
2364

    
2365
    if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2366
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
2367
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
2368
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
2369
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
2370
        for(i=2; i<18; i++){
2371
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
2372
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
2373
        }
2374
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
2375
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
2376
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
2377
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
2378
    }
2379
}
2380

    
2381
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){
2382
    MpegEncContext * const s = &h->s;
2383
    int temp8, i;
2384
    uint64_t temp64;
2385
    int deblock_left = (s->mb_x > 0);
2386
    int deblock_top  = (s->mb_y > 1);
2387

    
2388
    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);
2389

    
2390
    src_y  -= 2 *   linesize + 1;
2391
    src_cb -= 2 * uvlinesize + 1;
2392
    src_cr -= 2 * uvlinesize + 1;
2393

    
2394
#define XCHG(a,b,t,xchg)\
2395
t= a;\
2396
if(xchg)\
2397
    a= b;\
2398
b= t;
2399

    
2400
    if(deblock_left){
2401
        for(i = (!deblock_top)<<1; i<34; i++){
2402
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2403
        }
2404
    }
2405

    
2406
    if(deblock_top){
2407
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2408
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2409
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
2410
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
2411
        if(s->mb_x+1 < s->mb_width){
2412
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2413
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
2414
        }
2415
    }
2416

    
2417
    if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2418
        if(deblock_left){
2419
            for(i = (!deblock_top) << 1; i<18; i++){
2420
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
2421
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
2422
            }
2423
        }
2424
        if(deblock_top){
2425
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2426
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2427
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
2428
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
2429
        }
2430
    }
2431
}
2432

    
2433
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2434
    MpegEncContext * const s = &h->s;
2435
    const int mb_x= s->mb_x;
2436
    const int mb_y= s->mb_y;
2437
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2438
    const int mb_type= s->current_picture.mb_type[mb_xy];
2439
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2440
    int linesize, uvlinesize /*dct_offset*/;
2441
    int i;
2442
    int *block_offset = &h->block_offset[0];
2443
    const unsigned int bottom = mb_y & 1;
2444
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass), is_h264 = (simple || s->codec_id == CODEC_ID_H264);
2445
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2446
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2447

    
2448
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2449
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2450
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2451

    
2452
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2453
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2454

    
2455
    if (!simple && MB_FIELD) {
2456
        linesize   = h->mb_linesize   = s->linesize * 2;
2457
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2458
        block_offset = &h->block_offset[24];
2459
        if(mb_y&1){ //FIXME move out of this func?
2460
            dest_y -= s->linesize*15;
2461
            dest_cb-= s->uvlinesize*7;
2462
            dest_cr-= s->uvlinesize*7;
2463
        }
2464
        if(FRAME_MBAFF) {
2465
            int list;
2466
            for(list=0; list<h->list_count; list++){
2467
                if(!USES_LIST(mb_type, list))
2468
                    continue;
2469
                if(IS_16X16(mb_type)){
2470
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
2471
                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2472
                }else{
2473
                    for(i=0; i<16; i+=4){
2474
                        //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
2475
                        int ref = h->ref_cache[list][scan8[i]];
2476
                        if(ref >= 0)
2477
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2478
                    }
2479
                }
2480
            }
2481
        }
2482
    } else {
2483
        linesize   = h->mb_linesize   = s->linesize;
2484
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2485
//        dct_offset = s->linesize * 16;
2486
    }
2487

    
2488
    if(transform_bypass){
2489
        idct_dc_add =
2490
        idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2491
    }else if(IS_8x8DCT(mb_type)){
2492
        idct_dc_add = s->dsp.h264_idct8_dc_add;
2493
        idct_add = s->dsp.h264_idct8_add;
2494
    }else{
2495
        idct_dc_add = s->dsp.h264_idct_dc_add;
2496
        idct_add = s->dsp.h264_idct_add;
2497
    }
2498

    
2499
    if(!simple && FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
2500
       && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
2501
        int mbt_y = mb_y&~1;
2502
        uint8_t *top_y  = s->current_picture.data[0] + (mbt_y * 16* s->linesize  ) + mb_x * 16;
2503
        uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
2504
        uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
2505
        xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
2506
    }
2507

    
2508
    if (!simple && IS_INTRA_PCM(mb_type)) {
2509
        unsigned int x, y;
2510

    
2511
        // The pixels are stored in h->mb array in the same order as levels,
2512
        // copy them in output in the correct order.
2513
        for(i=0; i<16; i++) {
2514
            for (y=0; y<4; y++) {
2515
                for (x=0; x<4; x++) {
2516
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
2517
                }
2518
            }
2519
        }
2520
        for(i=16; i<16+4; i++) {
2521
            for (y=0; y<4; y++) {
2522
                for (x=0; x<4; x++) {
2523
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2524
                }
2525
            }
2526
        }
2527
        for(i=20; i<20+4; i++) {
2528
            for (y=0; y<4; y++) {
2529
                for (x=0; x<4; x++) {
2530
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2531
                }
2532
            }
2533
        }
2534
    } else {
2535
        if(IS_INTRA(mb_type)){
2536
            if(h->deblocking_filter && (simple || !FRAME_MBAFF))
2537
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2538

    
2539
            if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2540
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2541
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2542
            }
2543

    
2544
            if(IS_INTRA4x4(mb_type)){
2545
                if(simple || !s->encoding){
2546
                    if(IS_8x8DCT(mb_type)){
2547
                        for(i=0; i<16; i+=4){
2548
                            uint8_t * const ptr= dest_y + block_offset[i];
2549
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2550
                            const int nnz = h->non_zero_count_cache[ scan8[i] ];
2551
                            h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2552
                                                   (h->topright_samples_available<<i)&0x4000, linesize);
2553
                            if(nnz){
2554
                                if(nnz == 1 && h->mb[i*16])
2555
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
2556
                                else
2557
                                    idct_add(ptr, h->mb + i*16, linesize);
2558
                            }
2559
                        }
2560
                    }else
2561
                    for(i=0; i<16; i++){
2562
                        uint8_t * const ptr= dest_y + block_offset[i];
2563
                        uint8_t *topright;
2564
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2565
                        int nnz, tr;
2566

    
2567
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2568
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2569
                            assert(mb_y || linesize <= block_offset[i]);
2570
                            if(!topright_avail){
2571
                                tr= ptr[3 - linesize]*0x01010101;
2572
                                topright= (uint8_t*) &tr;
2573
                            }else
2574
                                topright= ptr + 4 - linesize;
2575
                        }else
2576
                            topright= NULL;
2577

    
2578
                        h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2579
                        nnz = h->non_zero_count_cache[ scan8[i] ];
2580
                        if(nnz){
2581
                            if(is_h264){
2582
                                if(nnz == 1 && h->mb[i*16])
2583
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
2584
                                else
2585
                                    idct_add(ptr, h->mb + i*16, linesize);
2586
                            }else
2587
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2588
                        }
2589
                    }
2590
                }
2591
            }else{
2592
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2593
                if(is_h264){
2594
                    if(!transform_bypass)
2595
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
2596
                }else
2597
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2598
            }
2599
            if(h->deblocking_filter && (simple || !FRAME_MBAFF))
2600
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2601
        }else if(is_h264){
2602
            hl_motion(h, dest_y, dest_cb, dest_cr,
2603
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2604
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2605
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2606
        }
2607

    
2608

    
2609
        if(!IS_INTRA4x4(mb_type)){
2610
            if(is_h264){
2611
                if(IS_INTRA16x16(mb_type)){
2612
                    for(i=0; i<16; i++){
2613
                        if(h->non_zero_count_cache[ scan8[i] ])
2614
                            idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2615
                        else if(h->mb[i*16])
2616
                            idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2617
                    }
2618
                }else{
2619
                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2620
                    for(i=0; i<16; i+=di){
2621
                        int nnz = h->non_zero_count_cache[ scan8[i] ];
2622
                        if(nnz){
2623
                            if(nnz==1 && h->mb[i*16])
2624
                                idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2625
                            else
2626
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2627
                        }
2628
                    }
2629
                }
2630
            }else{
2631
                for(i=0; i<16; i++){
2632
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2633
                        uint8_t * const ptr= dest_y + block_offset[i];
2634
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2635
                    }
2636
                }
2637
            }
2638
        }
2639

    
2640
        if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2641
            uint8_t *dest[2] = {dest_cb, dest_cr};
2642
            if(transform_bypass){
2643
                idct_add = idct_dc_add = s->dsp.add_pixels4;
2644
            }else{
2645
                idct_add = s->dsp.h264_idct_add;
2646
                idct_dc_add = s->dsp.h264_idct_dc_add;
2647
                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]);
2648
                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]);
2649
            }
2650
            if(is_h264){
2651
                for(i=16; i<16+8; i++){
2652
                    if(h->non_zero_count_cache[ scan8[i] ])
2653
                        idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2654
                    else if(h->mb[i*16])
2655
                        idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2656
                }
2657
            }else{
2658
                for(i=16; i<16+8; i++){
2659
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2660
                        uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
2661
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2662
                    }
2663
                }
2664
            }
2665
        }
2666
    }
2667
    if(h->deblocking_filter) {
2668
        if (!simple && FRAME_MBAFF) {
2669
            //FIXME try deblocking one mb at a time?
2670
            // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
2671
            const int mb_y = s->mb_y - 1;
2672
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
2673
            const int mb_xy= mb_x + mb_y*s->mb_stride;
2674
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
2675
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
2676
            if (!bottom) return;
2677
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2678
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2679
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2680

    
2681
            if(IS_INTRA(mb_type_top | mb_type_bottom))
2682
                xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
2683

    
2684
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
2685
            // deblock a pair
2686
            // top
2687
            s->mb_y--;
2688
            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);
2689
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
2690
            h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2691
            h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2692
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
2693
            // bottom
2694
            s->mb_y++;
2695
            tprintf(h->s.avctx, "call mbaff filter_mb\n");
2696
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
2697
            h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
2698
            h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
2699
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2700
        } else {
2701
            tprintf(h->s.avctx, "call filter_mb\n");
2702
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
2703
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
2704
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2705
        }
2706
    }
2707
}
2708

    
2709
/**
2710
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2711
 */
2712
static void hl_decode_mb_simple(H264Context *h){
2713
    hl_decode_mb_internal(h, 1);
2714
}
2715

    
2716
/**
2717
 * Process a macroblock; this handles edge cases, such as interlacing.
2718
 */
2719
static void av_noinline hl_decode_mb_complex(H264Context *h){
2720
    hl_decode_mb_internal(h, 0);
2721
}
2722

    
2723
static void hl_decode_mb(H264Context *h){
2724
    MpegEncContext * const s = &h->s;
2725
    const int mb_x= s->mb_x;
2726
    const int mb_y= s->mb_y;
2727
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2728
    const int mb_type= s->current_picture.mb_type[mb_xy];
2729
    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;
2730

    
2731
    if(!s->decode)
2732
        return;
2733

    
2734
    if (is_complex)
2735
        hl_decode_mb_complex(h);
2736
    else hl_decode_mb_simple(h);
2737
}
2738

    
2739
static void pic_as_field(Picture *pic, const int parity){
2740
    int i;
2741
    for (i = 0; i < 4; ++i) {
2742
        if (parity == PICT_BOTTOM_FIELD)
2743
            pic->data[i] += pic->linesize[i];
2744
        pic->reference = parity;
2745
        pic->linesize[i] *= 2;
2746
    }
2747
}
2748

    
2749
static int split_field_copy(Picture *dest, Picture *src,
2750
                            int parity, int id_add){
2751
    int match = !!(src->reference & parity);
2752

    
2753
    if (match) {
2754
        *dest = *src;
2755
        pic_as_field(dest, parity);
2756
        dest->pic_id *= 2;
2757
        dest->pic_id += id_add;
2758
    }
2759

    
2760
    return match;
2761
}
2762

    
2763
/**
2764
 * Split one reference list into field parts, interleaving by parity
2765
 * as per H.264 spec section 8.2.4.2.5. Output fields have their data pointers
2766
 * set to look at the actual start of data for that field.
2767
 *
2768
 * @param dest output list
2769
 * @param dest_len maximum number of fields to put in dest
2770
 * @param src the source reference list containing fields and/or field pairs
2771
 *            (aka short_ref/long_ref, or
2772
 *             refFrameListXShortTerm/refFrameListLongTerm in spec-speak)
2773
 * @param src_len number of Picture's in source (pairs and unmatched fields)
2774
 * @param parity the parity of the picture being decoded/needing
2775
 *        these ref pics (PICT_{TOP,BOTTOM}_FIELD)
2776
 * @return number of fields placed in dest
2777
 */
2778
static int split_field_half_ref_list(Picture *dest, int dest_len,
2779
                                     Picture *src,  int src_len,  int parity){
2780
    int same_parity   = 1;
2781
    int same_i        = 0;
2782
    int opp_i         = 0;
2783
    int out_i;
2784
    int field_output;
2785

    
2786
    for (out_i = 0; out_i < dest_len; out_i += field_output) {
2787
        if (same_parity && same_i < src_len) {
2788
            field_output = split_field_copy(dest + out_i, src + same_i,
2789
                                            parity, 1);
2790
            same_parity = !field_output;
2791
            same_i++;
2792

    
2793
        } else if (opp_i < src_len) {
2794
            field_output = split_field_copy(dest + out_i, src + opp_i,
2795
                                            PICT_FRAME - parity, 0);
2796
            same_parity = field_output;
2797
            opp_i++;
2798

    
2799
        } else {
2800
            break;
2801
        }
2802
    }
2803

    
2804
    return out_i;
2805
}
2806

    
2807
/**
2808
 * Split the reference frame list into a reference field list.
2809
 * This implements H.264 spec 8.2.4.2.5 for a combined input list.
2810
 * The input list contains both reference field pairs and
2811
 * unmatched reference fields; it is ordered as spec describes
2812
 * RefPicListX for frames in 8.2.4.2.1 and 8.2.4.2.3, except that
2813
 * unmatched field pairs are also present. Conceptually this is equivalent
2814
 * to concatenation of refFrameListXShortTerm with refFrameListLongTerm.
2815
 *
2816
 * @param dest output reference list where ordered fields are to be placed
2817
 * @param dest_len max number of fields to place at dest
2818
 * @param src source reference list, as described above
2819
 * @param src_len number of pictures (pairs and unmatched fields) in src
2820
 * @param parity parity of field being currently decoded
2821
 *        (one of PICT_{TOP,BOTTOM}_FIELD)
2822
 * @param long_i index into src array that holds first long reference picture,
2823
 *        or src_len if no long refs present.
2824
 */
2825
static int split_field_ref_list(Picture *dest, int dest_len,
2826
                                Picture *src,  int src_len,
2827
                                int parity,    int long_i){
2828

    
2829
    int i = split_field_half_ref_list(dest, dest_len, src, long_i, parity);
2830
    dest += i;
2831
    dest_len -= i;
2832

    
2833
    i += split_field_half_ref_list(dest, dest_len, src + long_i,
2834
                                   src_len - long_i, parity);
2835
    return i;
2836
}
2837

    
2838
/**
2839
 * fills the default_ref_list.
2840
 */
2841
static int fill_default_ref_list(H264Context *h){
2842
    MpegEncContext * const s = &h->s;
2843
    int i;
2844
    int smallest_poc_greater_than_current = -1;
2845
    int structure_sel;
2846
    Picture sorted_short_ref[32];
2847
    Picture field_entry_list[2][32];
2848
    Picture *frame_list[2];
2849

    
2850
    if (FIELD_PICTURE) {
2851
        structure_sel = PICT_FRAME;
2852
        frame_list[0] = field_entry_list[0];
2853
        frame_list[1] = field_entry_list[1];
2854
    } else {
2855
        structure_sel = 0;
2856
        frame_list[0] = h->default_ref_list[0];
2857
        frame_list[1] = h->default_ref_list[1];
2858
    }
2859

    
2860
    if(h->slice_type==B_TYPE){
2861
        int list;
2862
        int len[2];
2863
        int short_len[2];
2864
        int out_i;
2865
        int limit= INT_MIN;
2866

    
2867
        /* sort frame according to poc in B slice */
2868
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2869
            int best_i=INT_MIN;
2870
            int best_poc=INT_MAX;
2871

    
2872
            for(i=0; i<h->short_ref_count; i++){
2873
                const int poc= h->short_ref[i]->poc;
2874
                if(poc > limit && poc < best_poc){
2875
                    best_poc= poc;
2876
                    best_i= i;
2877
                }
2878
            }
2879

    
2880
            assert(best_i != INT_MIN);
2881

    
2882
            limit= best_poc;
2883
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2884
            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);
2885
            if (-1 == smallest_poc_greater_than_current) {
2886
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
2887
                    smallest_poc_greater_than_current = out_i;
2888
                }
2889
            }
2890
        }
2891

    
2892
        tprintf(h->s.avctx, "current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
2893

    
2894
        // find the largest poc
2895
        for(list=0; list<2; list++){
2896
            int index = 0;
2897
            int j= -99;
2898
            int step= list ? -1 : 1;
2899

    
2900
            for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
2901
                int sel;
2902
                while(j<0 || j>= h->short_ref_count){
2903
                    if(j != -99 && step == (list ? -1 : 1))
2904
                        return -1;
2905
                    step = -step;
2906
                    j= smallest_poc_greater_than_current + (step>>1);
2907
                }
2908
                sel = sorted_short_ref[j].reference | structure_sel;
2909
                if(sel != PICT_FRAME) continue;
2910
                frame_list[list][index  ]= sorted_short_ref[j];
2911
                frame_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
2912
            }
2913
            short_len[list] = index;
2914

    
2915
            for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
2916
                int sel;
2917
                if(h->long_ref[i] == NULL) continue;
2918
                sel = h->long_ref[i]->reference | structure_sel;
2919
                if(sel != PICT_FRAME) continue;
2920

    
2921
                frame_list[ list ][index  ]= *h->long_ref[i];
2922
                frame_list[ list ][index++].pic_id= i;;
2923
            }
2924
            len[list] = index;
2925

    
2926
            if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
2927
                // swap the two first elements of L1 when
2928
                // L0 and L1 are identical
2929
                Picture temp= frame_list[1][0];
2930
                frame_list[1][0] = frame_list[1][1];
2931
                frame_list[1][1] = temp;
2932
            }
2933

    
2934
        }
2935

    
2936
        for(list=0; list<2; list++){
2937
            if (FIELD_PICTURE)
2938
                len[list] = split_field_ref_list(h->default_ref_list[list],
2939
                                                 h->ref_count[list],
2940
                                                 frame_list[list],
2941
                                                 len[list],
2942
                                                 s->picture_structure,
2943
                                                 short_len[list]);
2944

    
2945
            if(len[list] < h->ref_count[ list ])
2946
                memset(&h->default_ref_list[list][len[list]], 0, sizeof(Picture)*(h->ref_count[ list ] - len[list]));
2947
        }
2948

    
2949

    
2950
    }else{
2951
        int index=0;
2952
        int short_len;
2953
        for(i=0; i<h->short_ref_count; i++){
2954
            int sel;
2955
            sel = h->short_ref[i]->reference | structure_sel;
2956
            if(sel != PICT_FRAME) continue;
2957
            frame_list[0][index  ]= *h->short_ref[i];
2958
            frame_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2959
        }
2960
        short_len = index;
2961
        for(i = 0; i < 16; i++){
2962
            int sel;
2963
            if(h->long_ref[i] == NULL) continue;
2964
            sel = h->long_ref[i]->reference | structure_sel;
2965
            if(sel != PICT_FRAME) continue;
2966
            frame_list[0][index  ]= *h->long_ref[i];
2967
            frame_list[0][index++].pic_id= i;;
2968
        }
2969

    
2970
        if (FIELD_PICTURE)
2971
            index = split_field_ref_list(h->default_ref_list[0],
2972
                                         h->ref_count[0], frame_list[0],
2973
                                         index, s->picture_structure,
2974
                                         short_len);
2975

    
2976
        if(index < h->ref_count[0])
2977
            memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2978
    }
2979
#ifdef TRACE
2980
    for (i=0; i<h->ref_count[0]; i++) {
2981
        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]);
2982
    }
2983
    if(h->slice_type==B_TYPE){
2984
        for (i=0; i<h->ref_count[1]; i++) {
2985
            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]);
2986
        }
2987
    }
2988
#endif
2989
    return 0;
2990
}
2991

    
2992
static void print_short_term(H264Context *h);
2993
static void print_long_term(H264Context *h);
2994

    
2995
/**
2996
 * Extract structure information about the picture described by pic_num in
2997
 * the current decoding context (frame or field). Note that pic_num is
2998
 * picture number without wrapping (so, 0<=pic_num<max_pic_num).
2999
 * @param pic_num picture number for which to extract structure information
3000
 * @param structure one of PICT_XXX describing structure of picture
3001
 *                      with pic_num
3002
 * @return frame number (short term) or long term index of picture
3003
 *         described by pic_num
3004
 */
3005
static int pic_num_extract(H264Context *h, int pic_num, int *structure){
3006
    MpegEncContext * const s = &h->s;
3007

    
3008
    *structure = s->picture_structure;
3009
    if(FIELD_PICTURE){
3010
        if (!(pic_num & 1))
3011
            /* opposite field */
3012
            *structure ^= PICT_FRAME;
3013
        pic_num >>= 1;
3014
    }
3015

    
3016
    return pic_num;
3017
}
3018

    
3019
static int decode_ref_pic_list_reordering(H264Context *h){
3020
    MpegEncContext * const s = &h->s;
3021
    int list, index, pic_structure;
3022

    
3023
    print_short_term(h);
3024
    print_long_term(h);
3025
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
3026

    
3027
    for(list=0; list<h->list_count; list++){
3028
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3029

    
3030
        if(get_bits1(&s->gb)){
3031
            int pred= h->curr_pic_num;
3032

    
3033
            for(index=0; ; index++){
3034
                unsigned int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3035
                unsigned int pic_id;
3036
                int i;
3037
                Picture *ref = NULL;
3038

    
3039
                if(reordering_of_pic_nums_idc==3)
3040
                    break;
3041

    
3042
                if(index >= h->ref_count[list]){
3043
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3044
                    return -1;
3045
                }
3046

    
3047
                if(reordering_of_pic_nums_idc<3){
3048
                    if(reordering_of_pic_nums_idc<2){
3049
                        const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3050
                        int frame_num;
3051

    
3052
                        if(abs_diff_pic_num > h->max_pic_num){
3053
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3054
                            return -1;
3055
                        }
3056

    
3057
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3058
                        else                                pred+= abs_diff_pic_num;
3059
                        pred &= h->max_pic_num - 1;
3060

    
3061
                        frame_num = pic_num_extract(h, pred, &pic_structure);
3062

    
3063
                        for(i= h->short_ref_count-1; i>=0; i--){
3064
                            ref = h->short_ref[i];
3065
                            assert(ref->reference);
3066
                            assert(!ref->long_ref);
3067
                            if(ref->data[0] != NULL &&
3068
                                   ref->frame_num == frame_num &&
3069
                                   (ref->reference & pic_structure) &&
3070
                                   ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3071
                                break;
3072
                        }
3073
                        if(i>=0)
3074
                            ref->pic_id= pred;
3075
                    }else{
3076
                        int long_idx;
3077
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3078

    
3079
                        long_idx= pic_num_extract(h, pic_id, &pic_structure);
3080

    
3081
                        if(long_idx>31){
3082
                            av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
3083
                            return -1;
3084
                        }
3085
                        ref = h->long_ref[long_idx];
3086
                        assert(!(ref && !ref->reference));
3087
                        if(ref && (ref->reference & pic_structure)){
3088
                            ref->pic_id= pic_id;
3089
                            assert(ref->long_ref);
3090
                            i=0;
3091
                        }else{
3092
                            i=-1;
3093
                        }
3094
                    }
3095

    
3096
                    if (i < 0) {
3097
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3098
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3099
                    } else {
3100
                        for(i=index; i+1<h->ref_count[list]; i++){
3101
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
3102
                                break;
3103
                        }
3104
                        for(; i > index; i--){
3105
                            h->ref_list[list][i]= h->ref_list[list][i-1];
3106
                        }
3107
                        h->ref_list[list][index]= *ref;
3108
                        if (FIELD_PICTURE){
3109
                            pic_as_field(&h->ref_list[list][index], pic_structure);
3110
                        }
3111
                    }
3112
                }else{
3113
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3114
                    return -1;
3115
                }
3116
            }
3117
        }
3118
    }
3119
    for(list=0; list<h->list_count; list++){
3120
        for(index= 0; index < h->ref_count[list]; index++){
3121
            if(!h->ref_list[list][index].data[0])
3122
                h->ref_list[list][index]= s->current_picture;
3123
        }
3124
    }
3125

    
3126
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3127
        direct_dist_scale_factor(h);
3128
    direct_ref_list_init(h);
3129
    return 0;
3130
}
3131

    
3132
static void fill_mbaff_ref_list(H264Context *h){
3133
    int list, i, j;
3134
    for(list=0; list<2; list++){ //FIXME try list_count
3135
        for(i=0; i<h->ref_count[list]; i++){
3136
            Picture *frame = &h->ref_list[list][i];
3137
            Picture *field = &h->ref_list[list][16+2*i];
3138
            field[0] = *frame;
3139
            for(j=0; j<3; j++)
3140
                field[0].linesize[j] <<= 1;
3141
            field[0].reference = PICT_TOP_FIELD;
3142
            field[1] = field[0];
3143
            for(j=0; j<3; j++)
3144
                field[1].data[j] += frame->linesize[j];
3145
            field[1].reference = PICT_BOTTOM_FIELD;
3146

    
3147
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
3148
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
3149
            for(j=0; j<2; j++){
3150
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
3151
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
3152
            }
3153
        }
3154
    }
3155
    for(j=0; j<h->ref_count[1]; j++){
3156
        for(i=0; i<h->ref_count[0]; i++)
3157
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
3158
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
3159
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
3160
    }
3161
}
3162

    
3163
static int pred_weight_table(H264Context *h){
3164
    MpegEncContext * const s = &h->s;
3165
    int list, i;
3166
    int luma_def, chroma_def;
3167

    
3168
    h->use_weight= 0;
3169
    h->use_weight_chroma= 0;
3170
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3171
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3172
    luma_def = 1<<h->luma_log2_weight_denom;
3173
    chroma_def = 1<<h->chroma_log2_weight_denom;
3174

    
3175
    for(list=0; list<2; list++){
3176
        for(i=0; i<h->ref_count[list]; i++){
3177
            int luma_weight_flag, chroma_weight_flag;
3178

    
3179
            luma_weight_flag= get_bits1(&s->gb);
3180
            if(luma_weight_flag){
3181
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3182
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3183
                if(   h->luma_weight[list][i] != luma_def
3184
                   || h->luma_offset[list][i] != 0)
3185
                    h->use_weight= 1;
3186
            }else{
3187
                h->luma_weight[list][i]= luma_def;
3188
                h->luma_offset[list][i]= 0;
3189
            }
3190

    
3191
            chroma_weight_flag= get_bits1(&s->gb);
3192
            if(chroma_weight_flag){
3193
                int j;
3194
                for(j=0; j<2; j++){
3195
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3196
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3197
                    if(   h->chroma_weight[list][i][j] != chroma_def
3198
                       || h->chroma_offset[list][i][j] != 0)
3199
                        h->use_weight_chroma= 1;
3200
                }
3201
            }else{
3202
                int j;
3203
                for(j=0; j<2; j++){
3204
                    h->chroma_weight[list][i][j]= chroma_def;
3205
                    h->chroma_offset[list][i][j]= 0;
3206
                }
3207
            }
3208
        }
3209
        if(h->slice_type != B_TYPE) break;
3210
    }
3211
    h->use_weight= h->use_weight || h->use_weight_chroma;
3212
    return 0;
3213
}
3214

    
3215
static void implicit_weight_table(H264Context *h){
3216
    MpegEncContext * const s = &h->s;
3217
    int ref0, ref1;
3218
    int cur_poc = s->current_picture_ptr->poc;
3219

    
3220
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3221
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3222
        h->use_weight= 0;
3223
        h->use_weight_chroma= 0;
3224
        return;
3225
    }
3226

    
3227
    h->use_weight= 2;
3228
    h->use_weight_chroma= 2;
3229
    h->luma_log2_weight_denom= 5;
3230
    h->chroma_log2_weight_denom= 5;
3231

    
3232
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3233
        int poc0 = h->ref_list[0][ref0].poc;
3234
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3235
            int poc1 = h->ref_list[1][ref1].poc;
3236
            int td = av_clip(poc1 - poc0, -128, 127);
3237
            if(td){
3238
                int tb = av_clip(cur_poc - poc0, -128, 127);
3239
                int tx = (16384 + (FFABS(td) >> 1)) / td;
3240
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3241
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3242
                    h->implicit_weight[ref0][ref1] = 32;
3243
                else
3244
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3245
            }else
3246
                h->implicit_weight[ref0][ref1] = 32;
3247
        }
3248
    }
3249
}
3250

    
3251
/**
3252
 * Mark a picture as no longer needed for reference. The refmask
3253
 * argument allows unreferencing of individual fields or the whole frame.
3254
 * If the picture becomes entirely unreferenced, but is being held for
3255
 * display purposes, it is marked as such.
3256
 * @param refmask mask of fields to unreference; the mask is bitwise
3257
 *                anded with the reference marking of pic
3258
 * @return non-zero if pic becomes entirely unreferenced (except possibly
3259
 *         for display purposes) zero if one of the fields remains in
3260
 *         reference
3261
 */
3262
static inline int unreference_pic(H264Context *h, Picture *pic, int refmask){
3263
    int i;
3264
    if (pic->reference &= refmask) {
3265
        return 0;
3266
    } else {
3267
        if(pic == h->delayed_output_pic)
3268
            pic->reference=DELAYED_PIC_REF;
3269
        else{
3270
            for(i = 0; h->delayed_pic[i]; i++)
3271
                if(pic == h->delayed_pic[i]){
3272
                    pic->reference=DELAYED_PIC_REF;
3273
                    break;
3274
                }
3275
        }
3276
        return 1;
3277
    }
3278
}
3279

    
3280
/**
3281
 * instantaneous decoder refresh.
3282
 */
3283
static void idr(H264Context *h){
3284
    int i;
3285

    
3286
    for(i=0; i<16; i++){
3287
        if (h->long_ref[i] != NULL) {
3288
            unreference_pic(h, h->long_ref[i], 0);
3289
            h->long_ref[i]= NULL;
3290
        }
3291
    }
3292
    h->long_ref_count=0;
3293

    
3294
    for(i=0; i<h->short_ref_count; i++){
3295
        unreference_pic(h, h->short_ref[i], 0);
3296
        h->short_ref[i]= NULL;
3297
    }
3298
    h->short_ref_count=0;
3299
}
3300

    
3301
/* forget old pics after a seek */
3302
static void flush_dpb(AVCodecContext *avctx){
3303
    H264Context *h= avctx->priv_data;
3304
    int i;
3305
    for(i=0; i<16; i++) {
3306
        if(h->delayed_pic[i])
3307
            h->delayed_pic[i]->reference= 0;
3308
        h->delayed_pic[i]= NULL;
3309
    }
3310
    if(h->delayed_output_pic)
3311
        h->delayed_output_pic->reference= 0;
3312
    h->delayed_output_pic= NULL;
3313
    idr(h);
3314
    if(h->s.current_picture_ptr)
3315
        h->s.current_picture_ptr->reference= 0;
3316
    h->s.first_field= 0;
3317
    ff_mpeg_flush(avctx);
3318
}
3319

    
3320
/**
3321
 * Find a Picture in the short term reference list by frame number.
3322
 * @param frame_num frame number to search for
3323
 * @param idx the index into h->short_ref where returned picture is found
3324
 *            undefined if no picture found.
3325
 * @return pointer to the found picture, or NULL if no pic with the provided
3326
 *                 frame number is found
3327
 */
3328
static Picture * find_short(H264Context *h, int frame_num, int *idx){
3329
    MpegEncContext * const s = &h->s;
3330
    int i;
3331

    
3332
    for(i=0; i<h->short_ref_count; i++){
3333
        Picture *pic= h->short_ref[i];
3334
        if(s->avctx->debug&FF_DEBUG_MMCO)
3335
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3336
        if(pic->frame_num == frame_num) {
3337
            *idx = i;
3338
            return pic;
3339
        }
3340
    }
3341
    return NULL;
3342
}
3343

    
3344
/**
3345
 * Remove a picture from the short term reference list by its index in
3346
 * that list.  This does no checking on the provided index; it is assumed
3347
 * to be valid. Other list entries are shifted down.
3348
 * @param i index into h->short_ref of picture to remove.
3349
 */
3350
static void remove_short_at_index(H264Context *h, int i){
3351
    assert(i > 0 && i < h->short_ref_count);
3352
    h->short_ref[i]= NULL;
3353
    if (--h->short_ref_count)
3354
        memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i)*sizeof(Picture*));
3355
}
3356

    
3357
/**
3358
 *
3359
 * @return the removed picture or NULL if an error occurs
3360
 */
3361
static Picture * remove_short(H264Context *h, int frame_num){
3362
    MpegEncContext * const s = &h->s;
3363
    Picture *pic;
3364
    int i;
3365

    
3366
    if(s->avctx->debug&FF_DEBUG_MMCO)
3367
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3368

    
3369
    pic = find_short(h, frame_num, &i);
3370
    if (pic)
3371
        remove_short_at_index(h, i);
3372

    
3373
    return pic;
3374
}
3375

    
3376
/**
3377
 * Remove a picture from the long term reference list by its index in
3378
 * that list.  This does no checking on the provided index; it is assumed
3379
 * to be valid. The removed entry is set to NULL. Other entries are unaffected.
3380
 * @param i index into h->long_ref of picture to remove.
3381
 */
3382
static void remove_long_at_index(H264Context *h, int i){
3383
    h->long_ref[i]= NULL;
3384
    h->long_ref_count--;
3385
}
3386

    
3387
/**
3388
 *
3389
 * @return the removed picture or NULL if an error occurs
3390
 */
3391
static Picture * remove_long(H264Context *h, int i){
3392
    Picture *pic;
3393

    
3394
    pic= h->long_ref[i];
3395
    if (pic)
3396
        remove_long_at_index(h, i);
3397

    
3398
    return pic;
3399
}
3400

    
3401
/**
3402
 * print short term list
3403
 */
3404
static void print_short_term(H264Context *h) {
3405
    uint32_t i;
3406
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3407
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3408
        for(i=0; i<h->short_ref_count; i++){
3409
            Picture *pic= h->short_ref[i];
3410
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3411
        }
3412
    }
3413
}
3414

    
3415
/**
3416
 * print long term list
3417
 */
3418
static void print_long_term(H264Context *h) {
3419
    uint32_t i;
3420
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3421
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3422
        for(i = 0; i < 16; i++){
3423
            Picture *pic= h->long_ref[i];
3424
            if (pic) {
3425
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3426
            }
3427
        }
3428
    }
3429
}
3430

    
3431
/**
3432
 * Executes the reference picture marking (memory management control operations).
3433
 */
3434
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3435
    MpegEncContext * const s = &h->s;
3436
    int i, j;
3437
    int current_ref_assigned=0;
3438
    Picture *pic;
3439

    
3440
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3441
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3442

    
3443
    for(i=0; i<mmco_count; i++){
3444
        int structure, frame_num, unref_pic;
3445
        if(s->avctx->debug&FF_DEBUG_MMCO)
3446
            av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_pic_num, h->mmco[i].long_arg);
3447

    
3448
        switch(mmco[i].opcode){
3449
        case MMCO_SHORT2UNUSED:
3450
            if(s->avctx->debug&FF_DEBUG_MMCO)
3451
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref short %d count %d\n", h->mmco[i].short_pic_num, h->short_ref_count);
3452
            frame_num = pic_num_extract(h, mmco[i].short_pic_num, &structure);
3453
            pic = find_short(h, frame_num, &j);
3454
            if (pic) {
3455
                if (unreference_pic(h, pic, structure ^ PICT_FRAME))
3456
                    remove_short_at_index(h, j);
3457
            } else if(s->avctx->debug&FF_DEBUG_MMCO)
3458
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref short failure\n");
3459
            break;
3460
        case MMCO_SHORT2LONG:
3461
            if (FIELD_PICTURE && mmco[i].long_arg < h->long_ref_count &&
3462
                    h->long_ref[mmco[i].long_arg]->frame_num ==
3463
                                              mmco[i].short_pic_num / 2) {
3464
                /* do nothing, we've already moved this field pair. */
3465
            } else {
3466
                int frame_num = mmco[i].short_pic_num >> FIELD_PICTURE;
3467

    
3468
                pic= remove_long(h, mmco[i].long_arg);
3469
                if(pic) unreference_pic(h, pic, 0);
3470

    
3471
                h->long_ref[ mmco[i].long_arg ]= remove_short(h, frame_num);
3472
                if (h->long_ref[ mmco[i].long_arg ]){
3473
                    h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3474
                    h->long_ref_count++;
3475
                }
3476
            }
3477
            break;
3478
        case MMCO_LONG2UNUSED:
3479
            j = pic_num_extract(h, mmco[i].long_arg, &structure);
3480
            pic = h->long_ref[j];
3481
            if (pic) {
3482
                if (unreference_pic(h, pic, structure ^ PICT_FRAME))
3483
                    remove_long_at_index(h, j);
3484
            } else if(s->avctx->debug&FF_DEBUG_MMCO)
3485
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref long failure\n");
3486
            break;
3487
        case MMCO_LONG:
3488
            unref_pic = 1;
3489
            if (FIELD_PICTURE && !s->first_field) {
3490
                if (h->long_ref[mmco[i].long_arg] == s->current_picture_ptr) {
3491
                    /* Just mark second field as referenced */
3492
                    unref_pic = 0;
3493
                } else if (s->current_picture_ptr->reference) {
3494
                    /* First field in pair is in short term list or
3495
                     * at a different long term index.
3496
                     * This is not allowed; see 7.4.3, notes 2 and 3.
3497
                     * Report the problem and keep the pair where it is,
3498
                     * and mark this field valid.
3499
                     */
3500
                    av_log(h->s.avctx, AV_LOG_ERROR,
3501
                        "illegal long term reference assignment for second "
3502
                        "field in complementary field pair (first field is "
3503
                        "short term or has non-matching long index)\n");
3504
                    unref_pic = 0;
3505
                }
3506
            }
3507

    
3508
            if (unref_pic) {
3509
                pic= remove_long(h, mmco[i].long_arg);
3510
                if(pic) unreference_pic(h, pic, 0);
3511

    
3512
                h->long_ref[ mmco[i].long_arg ]= s->current_picture_ptr;
3513
                h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3514
                h->long_ref_count++;
3515
            }
3516

    
3517
            s->current_picture_ptr->reference |= s->picture_structure;
3518
            current_ref_assigned=1;
3519
            break;
3520
        case MMCO_SET_MAX_LONG:
3521
            assert(mmco[i].long_arg <= 16);
3522
            // just remove the long term which index is greater than new max
3523
            for(j = mmco[i].long_arg; j<16; j++){
3524
                pic = remove_long(h, j);
3525
                if (pic) unreference_pic(h, pic, 0);
3526
            }
3527
            break;
3528
        case MMCO_RESET:
3529
            while(h->short_ref_count){
3530
                pic= remove_short(h, h->short_ref[0]->frame_num);
3531
                if(pic) unreference_pic(h, pic, 0);
3532
            }
3533
            for(j = 0; j < 16; j++) {
3534
                pic= remove_long(h, j);
3535
                if(pic) unreference_pic(h, pic, 0);
3536
            }
3537
            break;
3538
        default: assert(0);
3539
        }
3540
    }
3541

    
3542
    if (!current_ref_assigned && FIELD_PICTURE &&
3543
            !s->first_field && s->current_picture_ptr->reference) {
3544

    
3545
        /* Second field of complementary field pair; the first field of
3546
         * which is already referenced. If short referenced, it
3547
         * should be first entry in short_ref. If not, it must exist
3548
         * in long_ref; trying to put it on the short list here is an
3549
         * error in the encoded bit stream (ref: 7.4.3, NOTE 2 and 3).
3550
         */
3551
        if (h->short_ref_count && h->short_ref[0] == s->current_picture_ptr) {
3552
            /* Just mark the second field valid */
3553
            s->current_picture_ptr->reference = PICT_FRAME;
3554
        } else if (s->current_picture_ptr->long_ref) {
3555
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term reference "
3556
                                             "assignment for second field "
3557
                                             "in complementary field pair "
3558
                                             "(first field is long term)\n");
3559
        } else {
3560
            /*
3561
             * First field in reference, but not in any sensible place on our
3562
             * reference lists. This shouldn't happen unless reference
3563
             * handling somewhere else is wrong.
3564
             */
3565
            assert(0);
3566
        }
3567
        current_ref_assigned = 1;
3568
    }
3569

    
3570
    if(!current_ref_assigned){
3571
        pic= remove_short(h, s->current_picture_ptr->frame_num);
3572
        if(pic){
3573
            unreference_pic(h, pic, 0);
3574
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3575
        }
3576

    
3577
        if(h->short_ref_count)
3578
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3579

    
3580
        h->short_ref[0]= s->current_picture_ptr;
3581
        h->short_ref[0]->long_ref=0;
3582
        h->short_ref_count++;
3583
        s->current_picture_ptr->reference |= s->picture_structure;
3584
    }
3585

    
3586
    if (h->long_ref_count + h->short_ref_count > h->sps.ref_frame_count){
3587

    
3588
        /* We have too many reference frames, probably due to corrupted
3589
         * stream. Need to discard one frame. Prevents overrun of the
3590
         * short_ref and long_ref buffers.
3591
         */
3592
        av_log(h->s.avctx, AV_LOG_ERROR,
3593
               "number of reference frames exceeds max (probably "
3594
               "corrupt input), discarding one\n");
3595

    
3596
        if (h->long_ref_count) {
3597
            for (i = 0; i < 16; ++i)
3598
                if (h->long_ref[i])
3599
                    break;
3600

    
3601
            assert(i < 16);
3602
            pic = h->long_ref[i];
3603
            remove_long_at_index(h, i);
3604
        } else {
3605
            pic = h->short_ref[h->short_ref_count - 1];
3606
            remove_short_at_index(h, h->short_ref_count - 1);
3607
        }
3608
        unreference_pic(h, pic, 0);
3609
    }
3610

    
3611
    print_short_term(h);
3612
    print_long_term(h);
3613
    return 0;
3614
}
3615

    
3616
static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
3617
    MpegEncContext * const s = &h->s;
3618
    int i;
3619

    
3620
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3621
        s->broken_link= get_bits1(gb) -1;
3622
        h->mmco[0].long_arg= get_bits1(gb) - 1; // current_long_term_idx
3623
        if(h->mmco[0].long_arg == -1)
3624
            h->mmco_index= 0;
3625
        else{
3626
            h->mmco[0].opcode= MMCO_LONG;
3627
            h->mmco_index= 1;
3628
        }
3629
    }else{
3630
        if(get_bits1(gb)){ // adaptive_ref_pic_marking_mode_flag
3631
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
3632
                MMCOOpcode opcode= get_ue_golomb(gb);
3633

    
3634
                h->mmco[i].opcode= opcode;
3635
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3636
                    h->mmco[i].short_pic_num= (h->curr_pic_num - get_ue_golomb(gb) - 1) & (h->max_pic_num - 1);
3637
/*                    if(h->mmco[i].short_pic_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_pic_num ] == NULL){
3638
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
3639
                        return -1;
3640
                    }*/
3641
                }
3642
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3643
                    unsigned int long_arg= get_ue_golomb(gb);
3644
                    if(long_arg >= 32 || (long_arg >= 16 && !(opcode == MMCO_LONG2UNUSED && FIELD_PICTURE))){
3645
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3646
                        return -1;
3647
                    }
3648
                    h->mmco[i].long_arg= long_arg;
3649
                }
3650

    
3651
                if(opcode > (unsigned)MMCO_LONG){
3652
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3653
                    return -1;
3654
                }
3655
                if(opcode == MMCO_END)
3656
                    break;
3657
            }
3658
            h->mmco_index= i;
3659
        }else{
3660
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3661

    
3662
            if(h->short_ref_count && h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count &&
3663
                    !(FIELD_PICTURE && !s->first_field && s->current_picture_ptr->reference)) {
3664
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3665
                h->mmco[0].short_pic_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3666
                h->mmco_index= 1;
3667
                if (FIELD_PICTURE) {
3668
                    h->mmco[0].short_pic_num *= 2;
3669
                    h->mmco[1].opcode= MMCO_SHORT2UNUSED;
3670
                    h->mmco[1].short_pic_num= h->mmco[0].short_pic_num + 1;
3671
                    h->mmco_index= 2;
3672
                }
3673
            }else
3674
                h->mmco_index= 0;
3675
        }
3676
    }
3677

    
3678
    return 0;
3679
}
3680

    
3681
static int init_poc(H264Context *h){
3682
    MpegEncContext * const s = &h->s;
3683
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3684
    int field_poc[2];
3685

    
3686
    if(h->nal_unit_type == NAL_IDR_SLICE){
3687
        h->frame_num_offset= 0;
3688
    }else{
3689
        if(h->frame_num < h->prev_frame_num)
3690
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3691
        else
3692
            h->frame_num_offset= h->prev_frame_num_offset;
3693
    }
3694

    
3695
    if(h->sps.poc_type==0){
3696
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3697

    
3698
        if(h->nal_unit_type == NAL_IDR_SLICE){
3699
             h->prev_poc_msb=
3700
             h->prev_poc_lsb= 0;
3701
        }
3702

    
3703
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3704
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3705
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3706
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3707
        else
3708
            h->poc_msb = h->prev_poc_msb;
3709
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3710
        field_poc[0] =
3711
        field_poc[1] = h->poc_msb + h->poc_lsb;
3712
        if(s->picture_structure == PICT_FRAME)
3713
            field_poc[1] += h->delta_poc_bottom;
3714
    }else if(h->sps.poc_type==1){
3715
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3716
        int i;
3717

    
3718
        if(h->sps.poc_cycle_length != 0)
3719
            abs_frame_num = h->frame_num_offset + h->frame_num;
3720
        else
3721
            abs_frame_num = 0;
3722

    
3723
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
3724
            abs_frame_num--;
3725

    
3726
        expected_delta_per_poc_cycle = 0;
3727
        for(i=0; i < h->sps.poc_cycle_length; i++)
3728
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3729

    
3730
        if(abs_frame_num > 0){
3731
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3732
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3733

    
3734
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3735
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
3736
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3737
        } else
3738
            expectedpoc = 0;
3739

    
3740
        if(h->nal_ref_idc == 0)
3741
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3742

    
3743
        field_poc[0] = expectedpoc + h->delta_poc[0];
3744
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3745

    
3746
        if(s->picture_structure == PICT_FRAME)
3747
            field_poc[1] += h->delta_poc[1];
3748
    }else{
3749
        int poc;
3750
        if(h->nal_unit_type == NAL_IDR_SLICE){
3751
            poc= 0;
3752
        }else{
3753
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3754
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3755
        }
3756
        field_poc[0]= poc;
3757
        field_poc[1]= poc;
3758
    }
3759

    
3760
    if(s->picture_structure != PICT_BOTTOM_FIELD) {
3761
        s->current_picture_ptr->field_poc[0]= field_poc[0];
3762
        s->current_picture_ptr->poc = field_poc[0];
3763
    }
3764
    if(s->picture_structure != PICT_TOP_FIELD) {
3765
        s->current_picture_ptr->field_poc[1]= field_poc[1];
3766
        s->current_picture_ptr->poc = field_poc[1];
3767
    }
3768
    if(!FIELD_PICTURE || !s->first_field) {
3769
        Picture *cur = s->current_picture_ptr;
3770
        cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
3771
    }
3772

    
3773
    return 0;
3774
}
3775

    
3776

    
3777
/**
3778
 * initialize scan tables
3779
 */
3780
static void init_scan_tables(H264Context *h){
3781
    MpegEncContext * const s = &h->s;
3782
    int i;
3783
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
3784
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
3785
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
3786
    }else{
3787
        for(i=0; i<16; i++){
3788
#define T(x) (x>>2) | ((x<<2) & 0xF)
3789
            h->zigzag_scan[i] = T(zigzag_scan[i]);
3790
            h-> field_scan[i] = T( field_scan[i]);
3791
#undef T
3792
        }
3793
    }
3794
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
3795
        memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));
3796
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
3797
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
3798
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
3799
    }else{
3800
        for(i=0; i<64; i++){
3801
#define T(x) (x>>3) | ((x&7)<<3)
3802
            h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);
3803
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
3804
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
3805
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
3806
#undef T
3807
        }
3808
    }
3809
    if(h->sps.transform_bypass){ //FIXME same ugly
3810
        h->zigzag_scan_q0          = zigzag_scan;
3811
        h->zigzag_scan8x8_q0       = zigzag_scan8x8;
3812
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
3813
        h->field_scan_q0           = field_scan;
3814
        h->field_scan8x8_q0        = field_scan8x8;
3815
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
3816
    }else{
3817
        h->zigzag_scan_q0          = h->zigzag_scan;
3818
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
3819
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
3820
        h->field_scan_q0           = h->field_scan;
3821
        h->field_scan8x8_q0        = h->field_scan8x8;
3822
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
3823
    }
3824
}
3825

    
3826
/**
3827
 * Replicates H264 "master" context to thread contexts.
3828
 */
3829
static void clone_slice(H264Context *dst, H264Context *src)
3830
{
3831
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
3832
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
3833
    dst->s.current_picture      = src->s.current_picture;
3834
    dst->s.linesize             = src->s.linesize;
3835
    dst->s.uvlinesize           = src->s.uvlinesize;
3836
    dst->s.first_field          = src->s.first_field;
3837

    
3838
    dst->prev_poc_msb           = src->prev_poc_msb;
3839
    dst->prev_poc_lsb           = src->prev_poc_lsb;
3840
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
3841
    dst->prev_frame_num         = src->prev_frame_num;
3842
    dst->short_ref_count        = src->short_ref_count;
3843

    
3844
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
3845
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
3846
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
3847
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
3848

    
3849
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
3850
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
3851
}
3852

    
3853
/**
3854
 * decodes a slice header.
3855
 * this will allso call MPV_common_init() and frame_start() as needed
3856
 *
3857
 * @param h h264context
3858
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
3859
 *
3860
 * @return 0 if okay, <0 if an error occured, 1 if decoding must not be multithreaded
3861
 */
3862
static int decode_slice_header(H264Context *h, H264Context *h0){
3863
    MpegEncContext * const s = &h->s;
3864
    MpegEncContext * const s0 = &h0->s;
3865
    unsigned int first_mb_in_slice;
3866
    unsigned int pps_id;
3867
    int num_ref_idx_active_override_flag;
3868
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3869
    unsigned int slice_type, tmp, i;
3870
    int default_ref_list_done = 0;
3871
    int last_pic_structure;
3872

    
3873
    s->dropable= h->nal_ref_idc == 0;
3874

    
3875
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
3876
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
3877
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
3878
    }else{
3879
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
3880
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
3881
    }
3882

    
3883
    first_mb_in_slice= get_ue_golomb(&s->gb);
3884

    
3885
    if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
3886
        h0->current_slice = 0;
3887
        if (!s0->first_field)
3888
            s->current_picture_ptr= NULL;
3889
    }
3890

    
3891
    slice_type= get_ue_golomb(&s->gb);
3892
    if(slice_type > 9){
3893
        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);
3894
        return -1;
3895
    }
3896
    if(slice_type > 4){
3897
        slice_type -= 5;
3898
        h->slice_type_fixed=1;
3899
    }else
3900
        h->slice_type_fixed=0;
3901

    
3902
    slice_type= slice_type_map[ slice_type ];
3903
    if (slice_type == I_TYPE
3904
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
3905
        default_ref_list_done = 1;
3906
    }
3907
    h->slice_type= slice_type;
3908

    
3909
    s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
3910
    if (s->pict_type == B_TYPE && s0->last_picture_ptr == NULL) {
3911
        av_log(h->s.avctx, AV_LOG_ERROR,
3912
               "B picture before any references, skipping\n");
3913
        return -1;
3914
    }
3915

    
3916
    pps_id= get_ue_golomb(&s->gb);
3917
    if(pps_id>=MAX_PPS_COUNT){
3918
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3919
        return -1;
3920
    }
3921
    if(!h0->pps_buffers[pps_id]) {
3922
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3923
        return -1;
3924
    }
3925
    h->pps= *h0->pps_buffers[pps_id];
3926

    
3927
    if(!h0->sps_buffers[h->pps.sps_id]) {
3928
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3929
        return -1;
3930
    }
3931
    h->sps = *h0->sps_buffers[h->pps.sps_id];
3932

    
3933
    if(h == h0 && h->dequant_coeff_pps != pps_id){
3934
        h->dequant_coeff_pps = pps_id;
3935
        init_dequant_tables(h);
3936
    }
3937

    
3938
    s->mb_width= h->sps.mb_width;
3939
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3940

    
3941
    h->b_stride=  s->mb_width*4;
3942
    h->b8_stride= s->mb_width*2;
3943

    
3944
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3945
    if(h->sps.frame_mbs_only_flag)
3946
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
3947
    else
3948
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
3949

    
3950
    if (s->context_initialized
3951
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3952
        if(h != h0)
3953
            return -1;   // width / height changed during parallelized decoding
3954
        free_tables(h);
3955
        MPV_common_end(s);
3956
    }
3957
    if (!s->context_initialized) {
3958
        if(h != h0)
3959
            return -1;  // we cant (re-)initialize context during parallel decoding
3960
        if (MPV_common_init(s) < 0)
3961
            return -1;
3962
        s->first_field = 0;
3963

    
3964
        init_scan_tables(h);
3965
        alloc_tables(h);
3966

    
3967
        for(i = 1; i < s->avctx->thread_count; i++) {
3968
            H264Context *c;
3969
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
3970
            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
3971
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
3972
            c->sps = h->sps;
3973
            c->pps = h->pps;
3974
            init_scan_tables(c);
3975
            clone_tables(c, h);
3976
        }
3977

    
3978
        for(i = 0; i < s->avctx->thread_count; i++)
3979
            if(context_init(h->thread_context[i]) < 0)
3980
                return -1;
3981

    
3982
        s->avctx->width = s->width;
3983
        s->avctx->height = s->height;
3984
        s->avctx->sample_aspect_ratio= h->sps.sar;
3985
        if(!s->avctx->sample_aspect_ratio.den)
3986
            s->avctx->sample_aspect_ratio.den = 1;
3987

    
3988
        if(h->sps.timing_info_present_flag){
3989
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale};
3990
            if(h->x264_build > 0 && h->x264_build < 44)
3991
                s->avctx->time_base.den *= 2;
3992
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3993
                      s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
3994
        }
3995
    }
3996

    
3997
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3998

    
3999
    h->mb_mbaff = 0;
4000
    h->mb_aff_frame = 0;
4001
    last_pic_structure = s0->picture_structure;
4002
    if(h->sps.frame_mbs_only_flag){
4003
        s->picture_structure= PICT_FRAME;
4004
    }else{
4005
        if(get_bits1(&s->gb)) { //field_pic_flag
4006
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
4007
        } else {
4008
            s->picture_structure= PICT_FRAME;
4009
            h->mb_aff_frame = h->sps.mb_aff;
4010
        }
4011
    }
4012

    
4013
    if(h0->current_slice == 0){
4014
        /* See if we have a decoded first field looking for a pair... */
4015
        if (s0->first_field) {
4016
            assert(s0->current_picture_ptr);
4017
            assert(s0->current_picture_ptr->data[0]);
4018
            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
4019

    
4020
            /* figure out if we have a complementary field pair */
4021
            if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
4022
                /*
4023
                 * Previous field is unmatched. Don't display it, but let it
4024
                 * remain for reference if marked as such.
4025
                 */
4026
                s0->current_picture_ptr = NULL;
4027
                s0->first_field = FIELD_PICTURE;
4028

    
4029
            } else {
4030
                if (h->nal_ref_idc &&
4031
                        s0->current_picture_ptr->reference &&
4032
                        s0->current_picture_ptr->frame_num != h->frame_num) {
4033
                    /*
4034
                     * This and previous field were reference, but had
4035
                     * different frame_nums. Consider this field first in
4036
                     * pair. Throw away previous field except for reference
4037
                     * purposes.
4038
                     */
4039
                    s0->first_field = 1;
4040
                    s0->current_picture_ptr = NULL;
4041

    
4042
                } else {
4043
                    /* Second field in complementary pair */
4044
                    s0->first_field = 0;
4045
                }
4046
            }
4047

    
4048
        } else {
4049
            /* Frame or first field in a potentially complementary pair */
4050
            assert(!s0->current_picture_ptr);
4051
            s0->first_field = FIELD_PICTURE;
4052
        }
4053

    
4054
        if((!FIELD_PICTURE || s0->first_field) && frame_start(h) < 0) {
4055
            s0->first_field = 0;
4056
            return -1;
4057
        }
4058
    }
4059
    if(h != h0)
4060
        clone_slice(h, h0);
4061

    
4062
    s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
4063

    
4064
    assert(s->mb_num == s->mb_width * s->mb_height);
4065
    if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
4066
       first_mb_in_slice                    >= s->mb_num){
4067
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
4068
        return -1;
4069
    }
4070
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
4071
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
4072
    if (s->picture_structure == PICT_BOTTOM_FIELD)
4073
        s->resync_mb_y = s->mb_y = s->mb_y + 1;
4074
    assert(s->mb_y < s->mb_height);
4075

    
4076
    if(s->picture_structure==PICT_FRAME){
4077
        h->curr_pic_num=   h->frame_num;
4078
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
4079
    }else{
4080
        h->curr_pic_num= 2*h->frame_num + 1;
4081
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
4082
    }
4083

    
4084
    if(h->nal_unit_type == NAL_IDR_SLICE){
4085
        get_ue_golomb(&s->gb); /* idr_pic_id */
4086
    }
4087

    
4088
    if(h->sps.poc_type==0){
4089
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
4090

    
4091
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
4092
            h->delta_poc_bottom= get_se_golomb(&s->gb);
4093
        }
4094
    }
4095

    
4096
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
4097
        h->delta_poc[0]= get_se_golomb(&s->gb);
4098

    
4099
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
4100
            h->delta_poc[1]= get_se_golomb(&s->gb);
4101
    }
4102

    
4103
    init_poc(h);
4104

    
4105
    if(h->pps.redundant_pic_cnt_present){
4106
        h->redundant_pic_count= get_ue_golomb(&s->gb);
4107
    }
4108

    
4109
    //set defaults, might be overriden a few line later
4110
    h->ref_count[0]= h->pps.ref_count[0];
4111
    h->ref_count[1]= h->pps.ref_count[1];
4112

    
4113
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
4114
        if(h->slice_type == B_TYPE){
4115
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
4116
        }
4117
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
4118

    
4119
        if(num_ref_idx_active_override_flag){
4120
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
4121
            if(h->slice_type==B_TYPE)
4122
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
4123

    
4124
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
4125
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
4126
                h->ref_count[0]= h->ref_count[1]= 1;
4127
                return -1;
4128
            }
4129
        }
4130
        if(h->slice_type == B_TYPE)
4131
            h->list_count= 2;
4132
        else
4133
            h->list_count= 1;
4134
    }else
4135
        h->list_count= 0;
4136

    
4137
    if(!default_ref_list_done){
4138
        fill_default_ref_list(h);
4139
    }
4140

    
4141
    if(decode_ref_pic_list_reordering(h) < 0)
4142
        return -1;
4143

    
4144
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
4145
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
4146
        pred_weight_table(h);
4147
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
4148
        implicit_weight_table(h);
4149
    else
4150
        h->use_weight = 0;
4151

    
4152
    if(h->nal_ref_idc)
4153
        decode_ref_pic_marking(h0, &s->gb);
4154

    
4155
    if(FRAME_MBAFF)
4156
        fill_mbaff_ref_list(h);
4157

    
4158
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac ){
4159
        tmp = get_ue_golomb(&s->gb);
4160
        if(tmp > 2){
4161
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
4162
            return -1;
4163
        }
4164
        h->cabac_init_idc= tmp;
4165
    }
4166

    
4167
    h->last_qscale_diff = 0;
4168
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
4169
    if(tmp>51){
4170
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
4171
        return -1;
4172
    }
4173
    s->qscale= tmp;
4174
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
4175
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
4176
    //FIXME qscale / qp ... stuff
4177
    if(h->slice_type == SP_TYPE){
4178
        get_bits1(&s->gb); /* sp_for_switch_flag */
4179
    }
4180
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
4181
        get_se_golomb(&s->gb); /* slice_qs_delta */
4182
    }
4183

    
4184
    h->deblocking_filter = 1;
4185
    h->slice_alpha_c0_offset = 0;
4186
    h->slice_beta_offset = 0;
4187
    if( h->pps.deblocking_filter_parameters_present ) {
4188
        tmp= get_ue_golomb(&s->gb);
4189
        if(tmp > 2){
4190
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
4191
            return -1;
4192
        }
4193
        h->deblocking_filter= tmp;
4194
        if(h->deblocking_filter < 2)
4195
            h->deblocking_filter^= 1; // 1<->0
4196

    
4197
        if( h->deblocking_filter ) {
4198
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
4199
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
4200
        }
4201
    }
4202

    
4203
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
4204
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type != I_TYPE)
4205
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type == B_TYPE)
4206
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
4207
        h->deblocking_filter= 0;
4208

    
4209
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
4210
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
4211
            /* Cheat slightly for speed:
4212
               Do not bother to deblock across slices. */
4213
            h->deblocking_filter = 2;
4214
        } else {
4215
            h0->max_contexts = 1;
4216
            if(!h0->single_decode_warning) {
4217
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
4218
                h0->single_decode_warning = 1;
4219
            }
4220
            if(h != h0)
4221
                return 1; // deblocking switched inside frame
4222
        }
4223
    }
4224

    
4225
#if 0 //FMO
4226
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
4227
        slice_group_change_cycle= get_bits(&s->gb, ?);
4228
#endif
4229

    
4230
    h0->last_slice_type = slice_type;
4231
    h->slice_num = ++h0->current_slice;
4232

    
4233
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
4234
    h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
4235

    
4236
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4237
        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",
4238
               h->slice_num,
4239
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
4240
               first_mb_in_slice,
4241
               av_get_pict_type_char(h->slice_type),
4242
               pps_id, h->frame_num,
4243
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4244
               h->ref_count[0], h->ref_count[1],
4245
               s->qscale,
4246
               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4247
               h->use_weight,
4248
               h->use_weight==1 && h->use_weight_chroma ? "c" : ""
4249
               );
4250
    }
4251

    
4252
    return 0;
4253
}
4254

    
4255
/**
4256
 *
4257
 */
4258
static inline int get_level_prefix(GetBitContext *gb){
4259
    unsigned int buf;
4260
    int log;
4261

    
4262
    OPEN_READER(re, gb);
4263
    UPDATE_CACHE(re, gb);
4264
    buf=GET_CACHE(re, gb);
4265

    
4266
    log= 32 - av_log2(buf);
4267
#ifdef TRACE
4268
    print_bin(buf>>(32-log), log);
4269
    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__);
4270
#endif
4271

    
4272
    LAST_SKIP_BITS(re, gb, log);
4273
    CLOSE_READER(re, gb);
4274

    
4275
    return log-1;
4276
}
4277

    
4278
static inline int get_dct8x8_allowed(H264Context *h){
4279
    int i;
4280
    for(i=0; i<4; i++){
4281
        if(!IS_SUB_8X8(h->sub_mb_type[i])
4282
           || (!h->sps.direct_8x8_inference_flag && IS_DIRECT(h->sub_mb_type[i])))
4283
            return 0;
4284
    }
4285
    return 1;
4286
}
4287

    
4288
/**
4289
 * decodes a residual block.
4290
 * @param n block index
4291
 * @param scantable scantable
4292
 * @param max_coeff number of coefficients in the block
4293
 * @return <0 if an error occured
4294
 */
4295
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
4296
    MpegEncContext * const s = &h->s;
4297
    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};
4298
    int level[16];
4299
    int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;
4300

    
4301
    //FIXME put trailing_onex into the context
4302

    
4303
    if(n == CHROMA_DC_BLOCK_INDEX){
4304
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4305
        total_coeff= coeff_token>>2;
4306
    }else{
4307
        if(n == LUMA_DC_BLOCK_INDEX){
4308
            total_coeff= pred_non_zero_count(h, 0);
4309
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4310
            total_coeff= coeff_token>>2;
4311
        }else{
4312
            total_coeff= pred_non_zero_count(h, n);
4313
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4314
            total_coeff= coeff_token>>2;
4315
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4316
        }
4317
    }
4318

    
4319
    //FIXME set last_non_zero?
4320

    
4321
    if(total_coeff==0)
4322
        return 0;
4323
    if(total_coeff > (unsigned)max_coeff) {
4324
        av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
4325
        return -1;
4326
    }
4327

    
4328
    trailing_ones= coeff_token&3;
4329
    tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
4330
    assert(total_coeff<=16);
4331

    
4332
    for(i=0; i<trailing_ones; i++){
4333
        level[i]= 1 - 2*get_bits1(gb);
4334
    }
4335

    
4336
    if(i<total_coeff) {
4337
        int level_code, mask;
4338
        int suffix_length = total_coeff > 10 && trailing_ones < 3;
4339
        int prefix= get_level_prefix(gb);
4340

    
4341
        //first coefficient has suffix_length equal to 0 or 1
4342
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4343
            if(suffix_length)
4344
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4345
            else
4346
                level_code= (prefix<<suffix_length); //part
4347
        }else if(prefix==14){
4348
            if(suffix_length)
4349
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4350
            else
4351
                level_code= prefix + get_bits(gb, 4); //part
4352
        }else if(prefix==15){
4353
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
4354
            if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
4355
        }else{
4356
            av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
4357
            return -1;
4358
        }
4359

    
4360
        if(trailing_ones < 3) level_code += 2;
4361

    
4362
        suffix_length = 1;
4363
        if(level_code > 5)
4364
            suffix_length++;
4365
        mask= -(level_code&1);
4366
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
4367
        i++;
4368

    
4369
        //remaining coefficients have suffix_length > 0
4370
        for(;i<total_coeff;i++) {
4371
            static const int suffix_limit[7] = {0,5,11,23,47,95,INT_MAX };
4372
            prefix = get_level_prefix(gb);
4373
            if(prefix<15){
4374
                level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
4375
            }else if(prefix==15){
4376
                level_code =  (prefix<<suffix_length) + get_bits(gb, 12);
4377
            }else{
4378
                av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
4379
                return -1;
4380
            }
4381
            mask= -(level_code&1);
4382
            level[i]= (((2+level_code)>>1) ^ mask) - mask;
4383
            if(level_code > suffix_limit[suffix_length])
4384
                suffix_length++;
4385
        }
4386
    }
4387

    
4388
    if(total_coeff == max_coeff)
4389
        zeros_left=0;
4390
    else{
4391
        if(n == CHROMA_DC_BLOCK_INDEX)
4392
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4393
        else
4394
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4395
    }
4396

    
4397
    coeff_num = zeros_left + total_coeff - 1;
4398
    j = scantable[coeff_num];
4399
    if(n > 24){
4400
        block[j] = level[0];
4401
        for(i=1;i<total_coeff;i++) {
4402
            if(zeros_left <= 0)
4403
                run_before = 0;
4404
            else if(zeros_left < 7){
4405
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4406
            }else{
4407
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4408
            }
4409
            zeros_left -= run_before;
4410
            coeff_num -= 1 + run_before;
4411
            j= scantable[ coeff_num ];
4412

    
4413
            block[j]= level[i];
4414
        }
4415
    }else{
4416
        block[j] = (level[0] * qmul[j] + 32)>>6;
4417
        for(i=1;i<total_coeff;i++) {
4418
            if(zeros_left <= 0)
4419
                run_before = 0;
4420
            else if(zeros_left < 7){
4421
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4422
            }else{
4423
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4424
            }
4425
            zeros_left -= run_before;
4426
            coeff_num -= 1 + run_before;
4427
            j= scantable[ coeff_num ];
4428

    
4429
            block[j]= (level[i] * qmul[j] + 32)>>6;
4430
        }
4431
    }
4432

    
4433
    if(zeros_left<0){
4434
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4435
        return -1;
4436
    }
4437

    
4438
    return 0;
4439
}
4440

    
4441
static void predict_field_decoding_flag(H264Context *h){
4442
    MpegEncContext * const s = &h->s;
4443
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4444
    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
4445
                ? s->current_picture.mb_type[mb_xy-1]
4446
                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
4447
                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
4448
                : 0;
4449
    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4450
}
4451

    
4452
/**
4453
 * decodes a P_SKIP or B_SKIP macroblock
4454
 */
4455
static void decode_mb_skip(H264Context *h){
4456
    MpegEncContext * const s = &h->s;
4457
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4458
    int mb_type=0;
4459