Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 6f3c50f2

History | View | Annotate | Download (306 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
#ifdef ARCH_X86
39
#include "x86/h264_i386.h"
40
#endif
41

    
42
//#undef NDEBUG
43
#include <assert.h>
44

    
45
/**
46
 * Value of Picture.reference when Picture is not a reference picture, but
47
 * is held for delayed output.
48
 */
49
#define DELAYED_PIC_REF 4
50

    
51
static VLC coeff_token_vlc[4];
52
static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
53
static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
54

    
55
static VLC chroma_dc_coeff_token_vlc;
56
static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
57
static const int chroma_dc_coeff_token_vlc_table_size = 256;
58

    
59
static VLC total_zeros_vlc[15];
60
static VLC_TYPE total_zeros_vlc_tables[15][512][2];
61
static const int total_zeros_vlc_tables_size = 512;
62

    
63
static VLC chroma_dc_total_zeros_vlc[3];
64
static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
65
static const int chroma_dc_total_zeros_vlc_tables_size = 8;
66

    
67
static VLC run_vlc[6];
68
static VLC_TYPE run_vlc_tables[6][8][2];
69
static const int run_vlc_tables_size = 8;
70

    
71
static VLC run7_vlc;
72
static VLC_TYPE run7_vlc_table[96][2];
73
static const int run7_vlc_table_size = 96;
74

    
75
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
76
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
77
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);
78
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);
79
static Picture * remove_long(H264Context *h, int i, int ref_mask);
80

    
81
static av_always_inline uint32_t pack16to32(int a, int b){
82
#ifdef WORDS_BIGENDIAN
83
   return (b&0xFFFF) + (a<<16);
84
#else
85
   return (a&0xFFFF) + (b<<16);
86
#endif
87
}
88

    
89
static const uint8_t rem6[52]={
90
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,
91
};
92

    
93
static const uint8_t div6[52]={
94
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,
95
};
96

    
97
static const int left_block_options[4][8]={
98
    {0,1,2,3,7,10,8,11},
99
    {2,2,3,3,8,11,8,11},
100
    {0,0,1,1,7,10,7,10},
101
    {0,2,0,2,7,10,7,10}
102
};
103

    
104
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
105
    MpegEncContext * const s = &h->s;
106
    const int mb_xy= h->mb_xy;
107
    int topleft_xy, top_xy, topright_xy, left_xy[2];
108
    int topleft_type, top_type, topright_type, left_type[2];
109
    const int * left_block;
110
    int topleft_partition= -1;
111
    int i;
112

    
113
    top_xy     = mb_xy  - (s->mb_stride << FIELD_PICTURE);
114

    
115
    //FIXME deblocking could skip the intra and nnz parts.
116
    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
117
        return;
118

    
119
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
120
     * stuff, I can't imagine that these complex rules are worth it. */
121

    
122
    topleft_xy = top_xy - 1;
123
    topright_xy= top_xy + 1;
124
    left_xy[1] = left_xy[0] = mb_xy-1;
125
    left_block = left_block_options[0];
126
    if(FRAME_MBAFF){
127
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
128
        const int top_pair_xy      = pair_xy     - s->mb_stride;
129
        const int topleft_pair_xy  = top_pair_xy - 1;
130
        const int topright_pair_xy = top_pair_xy + 1;
131
        const int topleft_mb_field_flag  = IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
132
        const int top_mb_field_flag      = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
133
        const int topright_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
134
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
135
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
136
        const int bottom = (s->mb_y & 1);
137
        tprintf(s->avctx, "fill_caches: curr_mb_field_flag:%d, left_mb_field_flag:%d, topleft_mb_field_flag:%d, top_mb_field_flag:%d, topright_mb_field_flag:%d\n", curr_mb_field_flag, left_mb_field_flag, topleft_mb_field_flag, top_mb_field_flag, topright_mb_field_flag);
138

    
139
        if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
140
            top_xy -= s->mb_stride;
141
        }
142
        if (curr_mb_field_flag && (bottom || topleft_mb_field_flag)){
143
            topleft_xy -= s->mb_stride;
144
        } else if(bottom && !curr_mb_field_flag && left_mb_field_flag) {
145
            topleft_xy += s->mb_stride;
146
            // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
147
            topleft_partition = 0;
148
        }
149
        if (curr_mb_field_flag && (bottom || topright_mb_field_flag)){
150
            topright_xy -= s->mb_stride;
151
        }
152
        if (left_mb_field_flag != curr_mb_field_flag) {
153
            left_xy[1] = left_xy[0] = pair_xy - 1;
154
            if (curr_mb_field_flag) {
155
                left_xy[1] += s->mb_stride;
156
                left_block = left_block_options[3];
157
            } else {
158
                if (bottom) {
159
                    left_block = left_block_options[1];
160
                } else {
161
                    left_block= left_block_options[2];
162
                }
163
            }
164
        }
165
    }
166

    
167
    h->top_mb_xy = top_xy;
168
    h->left_mb_xy[0] = left_xy[0];
169
    h->left_mb_xy[1] = left_xy[1];
170
    if(for_deblock){
171
        topleft_type = 0;
172
        topright_type = 0;
173
        top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
174
        left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
175
        left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
176

    
177
        if(MB_MBAFF && !IS_INTRA(mb_type)){
178
            int list;
179
            for(list=0; list<h->list_count; list++){
180
                //These values where changed for ease of performing MC, we need to change them back
181
                //FIXME maybe we can make MC and loop filter use the same values or prevent
182
                //the MC code from changing ref_cache and rather use a temporary array.
183
                if(USES_LIST(mb_type,list)){
184
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
185
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
186
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
187
                    ref += h->b8_stride;
188
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
189
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
190
                }
191
            }
192
        }
193
    }else{
194
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
195
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
196
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
197
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
198
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
199

    
200
    if(IS_INTRA(mb_type)){
201
        int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
202
        h->topleft_samples_available=
203
        h->top_samples_available=
204
        h->left_samples_available= 0xFFFF;
205
        h->topright_samples_available= 0xEEEA;
206

    
207
        if(!(top_type & type_mask)){
208
            h->topleft_samples_available= 0xB3FF;
209
            h->top_samples_available= 0x33FF;
210
            h->topright_samples_available= 0x26EA;
211
        }
212
        if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
213
            if(IS_INTERLACED(mb_type)){
214
                if(!(left_type[0] & type_mask)){
215
                    h->topleft_samples_available&= 0xDFFF;
216
                    h->left_samples_available&= 0x5FFF;
217
                }
218
                if(!(left_type[1] & type_mask)){
219
                    h->topleft_samples_available&= 0xFF5F;
220
                    h->left_samples_available&= 0xFF5F;
221
                }
222
            }else{
223
                int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
224
                                ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
225
                assert(left_xy[0] == left_xy[1]);
226
                if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
227
                    h->topleft_samples_available&= 0xDF5F;
228
                    h->left_samples_available&= 0x5F5F;
229
                }
230
            }
231
        }else{
232
            if(!(left_type[0] & type_mask)){
233
                h->topleft_samples_available&= 0xDF5F;
234
                h->left_samples_available&= 0x5F5F;
235
            }
236
        }
237

    
238
        if(!(topleft_type & type_mask))
239
            h->topleft_samples_available&= 0x7FFF;
240

    
241
        if(!(topright_type & type_mask))
242
            h->topright_samples_available&= 0xFBFF;
243

    
244
        if(IS_INTRA4x4(mb_type)){
245
            if(IS_INTRA4x4(top_type)){
246
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
247
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
248
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
249
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
250
            }else{
251
                int pred;
252
                if(!(top_type & type_mask))
253
                    pred= -1;
254
                else{
255
                    pred= 2;
256
                }
257
                h->intra4x4_pred_mode_cache[4+8*0]=
258
                h->intra4x4_pred_mode_cache[5+8*0]=
259
                h->intra4x4_pred_mode_cache[6+8*0]=
260
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
261
            }
262
            for(i=0; i<2; i++){
263
                if(IS_INTRA4x4(left_type[i])){
264
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
265
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
266
                }else{
267
                    int pred;
268
                    if(!(left_type[i] & type_mask))
269
                        pred= -1;
270
                    else{
271
                        pred= 2;
272
                    }
273
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
274
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
275
                }
276
            }
277
        }
278
    }
279
    }
280

    
281

    
282
/*
283
0 . T T. T T T T
284
1 L . .L . . . .
285
2 L . .L . . . .
286
3 . T TL . . . .
287
4 L . .L . . . .
288
5 L . .. . . . .
289
*/
290
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
291
    if(top_type){
292
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
293
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
294
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
295
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
296

    
297
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
298
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
299

    
300
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
301
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
302

    
303
    }else{
304
        h->non_zero_count_cache[4+8*0]=
305
        h->non_zero_count_cache[5+8*0]=
306
        h->non_zero_count_cache[6+8*0]=
307
        h->non_zero_count_cache[7+8*0]=
308

    
309
        h->non_zero_count_cache[1+8*0]=
310
        h->non_zero_count_cache[2+8*0]=
311

    
312
        h->non_zero_count_cache[1+8*3]=
313
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
314

    
315
    }
316

    
317
    for (i=0; i<2; i++) {
318
        if(left_type[i]){
319
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
320
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
321
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
322
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
323
        }else{
324
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
325
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
326
            h->non_zero_count_cache[0+8*1 +   8*i]=
327
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
328
        }
329
    }
330

    
331
    if( h->pps.cabac ) {
332
        // top_cbp
333
        if(top_type) {
334
            h->top_cbp = h->cbp_table[top_xy];
335
        } else if(IS_INTRA(mb_type)) {
336
            h->top_cbp = 0x1C0;
337
        } else {
338
            h->top_cbp = 0;
339
        }
340
        // left_cbp
341
        if (left_type[0]) {
342
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
343
        } else if(IS_INTRA(mb_type)) {
344
            h->left_cbp = 0x1C0;
345
        } else {
346
            h->left_cbp = 0;
347
        }
348
        if (left_type[0]) {
349
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
350
        }
351
        if (left_type[1]) {
352
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
353
        }
354
    }
355

    
356
#if 1
357
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
358
        int list;
359
        for(list=0; list<h->list_count; list++){
360
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
361
                /*if(!h->mv_cache_clean[list]){
362
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
363
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
364
                    h->mv_cache_clean[list]= 1;
365
                }*/
366
                continue;
367
            }
368
            h->mv_cache_clean[list]= 0;
369

    
370
            if(USES_LIST(top_type, list)){
371
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
372
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
373
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
374
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
375
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
376
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
377
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
378
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
379
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
380
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
381
            }else{
382
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
383
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
384
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
385
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
386
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
387
            }
388

    
389
            for(i=0; i<2; i++){
390
                int cache_idx = scan8[0] - 1 + i*2*8;
391
                if(USES_LIST(left_type[i], list)){
392
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
393
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
394
                    *(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]];
395
                    *(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]];
396
                    h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
397
                    h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
398
                }else{
399
                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
400
                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
401
                    h->ref_cache[list][cache_idx  ]=
402
                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
403
                }
404
            }
405

    
406
            if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
407
                continue;
408

    
409
            if(USES_LIST(topleft_type, list)){
410
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
411
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
412
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
413
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
414
            }else{
415
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
416
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
417
            }
418

    
419
            if(USES_LIST(topright_type, list)){
420
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
421
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
422
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
423
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
424
            }else{
425
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
426
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
427
            }
428

    
429
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
430
                continue;
431

    
432
            h->ref_cache[list][scan8[5 ]+1] =
433
            h->ref_cache[list][scan8[7 ]+1] =
434
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
435
            h->ref_cache[list][scan8[4 ]] =
436
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
437
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
438
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
439
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
440
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
441
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
442

    
443
            if( h->pps.cabac ) {
444
                /* XXX beurk, Load mvd */
445
                if(USES_LIST(top_type, list)){
446
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
447
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
448
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
449
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
450
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
451
                }else{
452
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
453
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
454
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
455
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
456
                }
457
                if(USES_LIST(left_type[0], list)){
458
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
459
                    *(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]];
460
                    *(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]];
461
                }else{
462
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
463
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
464
                }
465
                if(USES_LIST(left_type[1], list)){
466
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
467
                    *(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]];
468
                    *(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]];
469
                }else{
470
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
471
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
472
                }
473
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
474
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
475
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
476
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
477
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
478

    
479
                if(h->slice_type_nos == FF_B_TYPE){
480
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
481

    
482
                    if(IS_DIRECT(top_type)){
483
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
484
                    }else if(IS_8X8(top_type)){
485
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
486
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
487
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
488
                    }else{
489
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
490
                    }
491

    
492
                    if(IS_DIRECT(left_type[0]))
493
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
494
                    else if(IS_8X8(left_type[0]))
495
                        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)];
496
                    else
497
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
498

    
499
                    if(IS_DIRECT(left_type[1]))
500
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
501
                    else if(IS_8X8(left_type[1]))
502
                        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)];
503
                    else
504
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
505
                }
506
            }
507

    
508
            if(FRAME_MBAFF){
509
#define MAP_MVS\
510
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
511
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
512
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
513
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
514
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
515
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
516
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
517
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
518
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
519
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
520
                if(MB_FIELD){
521
#define MAP_F2F(idx, mb_type)\
522
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
523
                        h->ref_cache[list][idx] <<= 1;\
524
                        h->mv_cache[list][idx][1] /= 2;\
525
                        h->mvd_cache[list][idx][1] /= 2;\
526
                    }
527
                    MAP_MVS
528
#undef MAP_F2F
529
                }else{
530
#define MAP_F2F(idx, mb_type)\
531
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
532
                        h->ref_cache[list][idx] >>= 1;\
533
                        h->mv_cache[list][idx][1] <<= 1;\
534
                        h->mvd_cache[list][idx][1] <<= 1;\
535
                    }
536
                    MAP_MVS
537
#undef MAP_F2F
538
                }
539
            }
540
        }
541
    }
542
#endif
543

    
544
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
545
}
546

    
547
static inline void write_back_intra_pred_mode(H264Context *h){
548
    const int mb_xy= h->mb_xy;
549

    
550
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
551
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
552
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
553
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
554
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
555
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
556
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
557
}
558

    
559
/**
560
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
561
 */
562
static inline int check_intra4x4_pred_mode(H264Context *h){
563
    MpegEncContext * const s = &h->s;
564
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
565
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
566
    int i;
567

    
568
    if(!(h->top_samples_available&0x8000)){
569
        for(i=0; i<4; i++){
570
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
571
            if(status<0){
572
                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);
573
                return -1;
574
            } else if(status){
575
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
576
            }
577
        }
578
    }
579

    
580
    if((h->left_samples_available&0x8888)!=0x8888){
581
        static const int mask[4]={0x8000,0x2000,0x80,0x20};
582
        for(i=0; i<4; i++){
583
            if(!(h->left_samples_available&mask[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

    
595
    return 0;
596
} //FIXME cleanup like next
597

    
598
/**
599
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
600
 */
601
static inline int check_intra_pred_mode(H264Context *h, int mode){
602
    MpegEncContext * const s = &h->s;
603
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
604
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
605

    
606
    if(mode > 6U) {
607
        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);
608
        return -1;
609
    }
610

    
611
    if(!(h->top_samples_available&0x8000)){
612
        mode= top[ mode ];
613
        if(mode<0){
614
            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);
615
            return -1;
616
        }
617
    }
618

    
619
    if((h->left_samples_available&0x8080) != 0x8080){
620
        mode= left[ mode ];
621
        if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
622
            mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
623
        }
624
        if(mode<0){
625
            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);
626
            return -1;
627
        }
628
    }
629

    
630
    return mode;
631
}
632

    
633
/**
634
 * gets the predicted intra4x4 prediction mode.
635
 */
636
static inline int pred_intra_mode(H264Context *h, int n){
637
    const int index8= scan8[n];
638
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
639
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
640
    const int min= FFMIN(left, top);
641

    
642
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
643

    
644
    if(min<0) return DC_PRED;
645
    else      return min;
646
}
647

    
648
static inline void write_back_non_zero_count(H264Context *h){
649
    const int mb_xy= h->mb_xy;
650

    
651
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
652
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
653
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
654
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
655
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
656
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
657
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
658

    
659
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
660
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
661
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
662

    
663
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
664
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
665
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
666
}
667

    
668
/**
669
 * gets the predicted number of non-zero coefficients.
670
 * @param n block index
671
 */
672
static inline int pred_non_zero_count(H264Context *h, int n){
673
    const int index8= scan8[n];
674
    const int left= h->non_zero_count_cache[index8 - 1];
675
    const int top = h->non_zero_count_cache[index8 - 8];
676
    int i= left + top;
677

    
678
    if(i<64) i= (i+1)>>1;
679

    
680
    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
681

    
682
    return i&31;
683
}
684

    
685
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
686
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
687
    MpegEncContext *s = &h->s;
688

    
689
    /* there is no consistent mapping of mvs to neighboring locations that will
690
     * make mbaff happy, so we can't move all this logic to fill_caches */
691
    if(FRAME_MBAFF){
692
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
693
        const int16_t *mv;
694
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
695
        *C = h->mv_cache[list][scan8[0]-2];
696

    
697
        if(!MB_FIELD
698
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
699
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
700
            if(IS_INTERLACED(mb_types[topright_xy])){
701
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
702
                const int x4 = X4, y4 = Y4;\
703
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
704
                if(!USES_LIST(mb_type,list))\
705
                    return LIST_NOT_USED;\
706
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
707
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
708
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
709
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
710

    
711
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
712
            }
713
        }
714
        if(topright_ref == PART_NOT_AVAILABLE
715
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
716
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
717
            if(!MB_FIELD
718
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
719
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
720
            }
721
            if(MB_FIELD
722
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
723
               && i >= scan8[0]+8){
724
                // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
725
                SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
726
            }
727
        }
728
#undef SET_DIAG_MV
729
    }
730

    
731
    if(topright_ref != PART_NOT_AVAILABLE){
732
        *C= h->mv_cache[list][ i - 8 + part_width ];
733
        return topright_ref;
734
    }else{
735
        tprintf(s->avctx, "topright MV not available\n");
736

    
737
        *C= h->mv_cache[list][ i - 8 - 1 ];
738
        return h->ref_cache[list][ i - 8 - 1 ];
739
    }
740
}
741

    
742
/**
743
 * gets the predicted MV.
744
 * @param n the block index
745
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
746
 * @param mx the x component of the predicted motion vector
747
 * @param my the y component of the predicted motion vector
748
 */
749
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
750
    const int index8= scan8[n];
751
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
752
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
753
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
754
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
755
    const int16_t * C;
756
    int diagonal_ref, match_count;
757

    
758
    assert(part_width==1 || part_width==2 || part_width==4);
759

    
760
/* mv_cache
761
  B . . A T T T T
762
  U . . L . . , .
763
  U . . L . . . .
764
  U . . L . . , .
765
  . . . L . . . .
766
*/
767

    
768
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
769
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
770
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
771
    if(match_count > 1){ //most common
772
        *mx= mid_pred(A[0], B[0], C[0]);
773
        *my= mid_pred(A[1], B[1], C[1]);
774
    }else if(match_count==1){
775
        if(left_ref==ref){
776
            *mx= A[0];
777
            *my= A[1];
778
        }else if(top_ref==ref){
779
            *mx= B[0];
780
            *my= B[1];
781
        }else{
782
            *mx= C[0];
783
            *my= C[1];
784
        }
785
    }else{
786
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
787
            *mx= A[0];
788
            *my= A[1];
789
        }else{
790
            *mx= mid_pred(A[0], B[0], C[0]);
791
            *my= mid_pred(A[1], B[1], C[1]);
792
        }
793
    }
794

    
795
    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);
796
}
797

    
798
/**
799
 * gets the directionally predicted 16x8 MV.
800
 * @param n the block index
801
 * @param mx the x component of the predicted motion vector
802
 * @param my the y component of the predicted motion vector
803
 */
804
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
805
    if(n==0){
806
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
807
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
808

    
809
        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);
810

    
811
        if(top_ref == ref){
812
            *mx= B[0];
813
            *my= B[1];
814
            return;
815
        }
816
    }else{
817
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
818
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
819

    
820
        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);
821

    
822
        if(left_ref == ref){
823
            *mx= A[0];
824
            *my= A[1];
825
            return;
826
        }
827
    }
828

    
829
    //RARE
830
    pred_motion(h, n, 4, list, ref, mx, my);
831
}
832

    
833
/**
834
 * gets the directionally predicted 8x16 MV.
835
 * @param n the block index
836
 * @param mx the x component of the predicted motion vector
837
 * @param my the y component of the predicted motion vector
838
 */
839
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
840
    if(n==0){
841
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
842
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
843

    
844
        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);
845

    
846
        if(left_ref == ref){
847
            *mx= A[0];
848
            *my= A[1];
849
            return;
850
        }
851
    }else{
852
        const int16_t * C;
853
        int diagonal_ref;
854

    
855
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
856

    
857
        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);
858

    
859
        if(diagonal_ref == ref){
860
            *mx= C[0];
861
            *my= C[1];
862
            return;
863
        }
864
    }
865

    
866
    //RARE
867
    pred_motion(h, n, 2, list, ref, mx, my);
868
}
869

    
870
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
871
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
872
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
873

    
874
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
875

    
876
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
877
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
878
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
879

    
880
        *mx = *my = 0;
881
        return;
882
    }
883

    
884
    pred_motion(h, 0, 4, 0, 0, mx, my);
885

    
886
    return;
887
}
888

    
889
static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
890
    int poc0 = h->ref_list[0][i].poc;
891
    int td = av_clip(poc1 - poc0, -128, 127);
892
    if(td == 0 || h->ref_list[0][i].long_ref){
893
        return 256;
894
    }else{
895
        int tb = av_clip(poc - poc0, -128, 127);
896
        int tx = (16384 + (FFABS(td) >> 1)) / td;
897
        return av_clip((tb*tx + 32) >> 6, -1024, 1023);
898
    }
899
}
900

    
901
static inline void direct_dist_scale_factor(H264Context * const h){
902
    MpegEncContext * const s = &h->s;
903
    const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
904
    const int poc1 = h->ref_list[1][0].poc;
905
    int i, field;
906
    for(field=0; field<2; field++){
907
        const int poc  = h->s.current_picture_ptr->field_poc[field];
908
        const int poc1 = h->ref_list[1][0].field_poc[field];
909
        for(i=0; i < 2*h->ref_count[0]; i++)
910
            h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
911
    }
912

    
913
    for(i=0; i<h->ref_count[0]; i++){
914
        h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
915
    }
916
}
917

    
918
static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
919
    MpegEncContext * const s = &h->s;
920
    Picture * const ref1 = &h->ref_list[1][0];
921
    int j, old_ref, rfield;
922
    int start= mbafi ? 16                      : 0;
923
    int end  = mbafi ? 16+2*h->ref_count[list] : h->ref_count[list];
924
    int interl= mbafi || s->picture_structure != PICT_FRAME;
925

    
926
    /* bogus; fills in for missing frames */
927
    memset(map[list], 0, sizeof(map[list]));
928

    
929
    for(rfield=0; rfield<2; rfield++){
930
        for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
931
            int poc = ref1->ref_poc[colfield][list][old_ref];
932

    
933
            if     (!interl)
934
                poc |= 3;
935
            else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
936
                poc= (poc&~3) + rfield + 1;
937

    
938
            for(j=start; j<end; j++){
939
                if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
940
                    int cur_ref= mbafi ? (j-16)^field : j;
941
                    map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
942
                    if(rfield == field)
943
                        map[list][old_ref] = cur_ref;
944
                    break;
945
                }
946
            }
947
        }
948
    }
949
}
950

    
951
static inline void direct_ref_list_init(H264Context * const h){
952
    MpegEncContext * const s = &h->s;
953
    Picture * const ref1 = &h->ref_list[1][0];
954
    Picture * const cur = s->current_picture_ptr;
955
    int list, j, field;
956
    int sidx= (s->picture_structure&1)^1;
957
    int ref1sidx= (ref1->reference&1)^1;
958

    
959
    for(list=0; list<2; list++){
960
        cur->ref_count[sidx][list] = h->ref_count[list];
961
        for(j=0; j<h->ref_count[list]; j++)
962
            cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
963
    }
964

    
965
    if(s->picture_structure == PICT_FRAME){
966
        memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
967
        memcpy(cur->ref_poc  [1], cur->ref_poc  [0], sizeof(cur->ref_poc  [0]));
968
    }
969

    
970
    cur->mbaff= FRAME_MBAFF;
971

    
972
    if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
973
        return;
974

    
975
    for(list=0; list<2; list++){
976
        fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
977
        for(field=0; field<2; field++)
978
            fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
979
    }
980
}
981

    
982
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
983
    MpegEncContext * const s = &h->s;
984
    int b8_stride = h->b8_stride;
985
    int b4_stride = h->b_stride;
986
    int mb_xy = h->mb_xy;
987
    int mb_type_col[2];
988
    const int16_t (*l1mv0)[2], (*l1mv1)[2];
989
    const int8_t *l1ref0, *l1ref1;
990
    const int is_b8x8 = IS_8X8(*mb_type);
991
    unsigned int sub_mb_type;
992
    int i8, i4;
993

    
994
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
995

    
996
    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
997
        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
998
            int cur_poc = s->current_picture_ptr->poc;
999
            int *col_poc = h->ref_list[1]->field_poc;
1000
            int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
1001
            mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
1002
            b8_stride = 0;
1003
        }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity
1004
            int fieldoff= 2*(h->ref_list[1][0].reference)-3;
1005
            mb_xy += s->mb_stride*fieldoff;
1006
        }
1007
        goto single_col;
1008
    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
1009
        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
1010
            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
1011
            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
1012
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
1013
            b8_stride *= 3;
1014
            b4_stride *= 6;
1015
            //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag
1016
            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
1017
                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
1018
                && !is_b8x8){
1019
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1020
                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
1021
            }else{
1022
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1023
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1024
            }
1025
        }else{                                           //     AFR/FR    -> AFR/FR
1026
single_col:
1027
            mb_type_col[0] =
1028
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
1029
            if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
1030
                /* FIXME save sub mb types from previous frames (or derive from MVs)
1031
                * so we know exactly what block size to use */
1032
                sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1033
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1034
            }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
1035
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1036
                *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1037
            }else{
1038
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1039
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1040
            }
1041
        }
1042
    }
1043

    
1044
    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
1045
    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
1046
    l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
1047
    l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
1048
    if(!b8_stride){
1049
        if(s->mb_y&1){
1050
            l1ref0 += h->b8_stride;
1051
            l1ref1 += h->b8_stride;
1052
            l1mv0  +=  2*b4_stride;
1053
            l1mv1  +=  2*b4_stride;
1054
        }
1055
    }
1056

    
1057
    if(h->direct_spatial_mv_pred){
1058
        int ref[2];
1059
        int mv[2][2];
1060
        int list;
1061

    
1062
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1063

    
1064
        /* ref = min(neighbors) */
1065
        for(list=0; list<2; list++){
1066
            int refa = h->ref_cache[list][scan8[0] - 1];
1067
            int refb = h->ref_cache[list][scan8[0] - 8];
1068
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1069
            if(refc == PART_NOT_AVAILABLE)
1070
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1071
            ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
1072
            if(ref[list] < 0)
1073
                ref[list] = -1;
1074
        }
1075

    
1076
        if(ref[0] < 0 && ref[1] < 0){
1077
            ref[0] = ref[1] = 0;
1078
            mv[0][0] = mv[0][1] =
1079
            mv[1][0] = mv[1][1] = 0;
1080
        }else{
1081
            for(list=0; list<2; list++){
1082
                if(ref[list] >= 0)
1083
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1084
                else
1085
                    mv[list][0] = mv[list][1] = 0;
1086
            }
1087
        }
1088

    
1089
        if(ref[1] < 0){
1090
            if(!is_b8x8)
1091
                *mb_type &= ~MB_TYPE_L1;
1092
            sub_mb_type &= ~MB_TYPE_L1;
1093
        }else if(ref[0] < 0){
1094
            if(!is_b8x8)
1095
                *mb_type &= ~MB_TYPE_L0;
1096
            sub_mb_type &= ~MB_TYPE_L0;
1097
        }
1098

    
1099
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1100
            for(i8=0; i8<4; i8++){
1101
                int x8 = i8&1;
1102
                int y8 = i8>>1;
1103
                int xy8 = x8+y8*b8_stride;
1104
                int xy4 = 3*x8+y8*b4_stride;
1105
                int a=0, b=0;
1106

    
1107
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1108
                    continue;
1109
                h->sub_mb_type[i8] = sub_mb_type;
1110

    
1111
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1112
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1113
                if(!IS_INTRA(mb_type_col[y8])
1114
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1115
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1116
                    if(ref[0] > 0)
1117
                        a= pack16to32(mv[0][0],mv[0][1]);
1118
                    if(ref[1] > 0)
1119
                        b= pack16to32(mv[1][0],mv[1][1]);
1120
                }else{
1121
                    a= pack16to32(mv[0][0],mv[0][1]);
1122
                    b= pack16to32(mv[1][0],mv[1][1]);
1123
                }
1124
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1125
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1126
            }
1127
        }else if(IS_16X16(*mb_type)){
1128
            int a=0, b=0;
1129

    
1130
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1131
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1132
            if(!IS_INTRA(mb_type_col[0])
1133
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1134
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1135
                       && (h->x264_build>33 || !h->x264_build)))){
1136
                if(ref[0] > 0)
1137
                    a= pack16to32(mv[0][0],mv[0][1]);
1138
                if(ref[1] > 0)
1139
                    b= pack16to32(mv[1][0],mv[1][1]);
1140
            }else{
1141
                a= pack16to32(mv[0][0],mv[0][1]);
1142
                b= pack16to32(mv[1][0],mv[1][1]);
1143
            }
1144
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1145
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1146
        }else{
1147
            for(i8=0; i8<4; i8++){
1148
                const int x8 = i8&1;
1149
                const int y8 = i8>>1;
1150

    
1151
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1152
                    continue;
1153
                h->sub_mb_type[i8] = sub_mb_type;
1154

    
1155
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1156
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1157
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1158
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1159

    
1160
                /* col_zero_flag */
1161
                if(!IS_INTRA(mb_type_col[0]) && (   l1ref0[x8 + y8*b8_stride] == 0
1162
                                              || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
1163
                                                  && (h->x264_build>33 || !h->x264_build)))){
1164
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
1165
                    if(IS_SUB_8X8(sub_mb_type)){
1166
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1167
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1168
                            if(ref[0] == 0)
1169
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1170
                            if(ref[1] == 0)
1171
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1172
                        }
1173
                    }else
1174
                    for(i4=0; i4<4; i4++){
1175
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1176
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1177
                            if(ref[0] == 0)
1178
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1179
                            if(ref[1] == 0)
1180
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1181
                        }
1182
                    }
1183
                }
1184
            }
1185
        }
1186
    }else{ /* direct temporal mv pred */
1187
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1188
        const int *dist_scale_factor = h->dist_scale_factor;
1189
        int ref_offset= 0;
1190

    
1191
        if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
1192
            map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
1193
            map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
1194
            dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];
1195
        }
1196
        if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
1197
            ref_offset += 16;
1198

    
1199
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1200
            /* FIXME assumes direct_8x8_inference == 1 */
1201
            int y_shift  = 2*!IS_INTERLACED(*mb_type);
1202

    
1203
            for(i8=0; i8<4; i8++){
1204
                const int x8 = i8&1;
1205
                const int y8 = i8>>1;
1206
                int ref0, scale;
1207
                const int16_t (*l1mv)[2]= l1mv0;
1208

    
1209
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1210
                    continue;
1211
                h->sub_mb_type[i8] = sub_mb_type;
1212

    
1213
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1214
                if(IS_INTRA(mb_type_col[y8])){
1215
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1216
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1217
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1218
                    continue;
1219
                }
1220

    
1221
                ref0 = l1ref0[x8 + y8*b8_stride];
1222
                if(ref0 >= 0)
1223
                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
1224
                else{
1225
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1226
                    l1mv= l1mv1;
1227
                }
1228
                scale = dist_scale_factor[ref0];
1229
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1230

    
1231
                {
1232
                    const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
1233
                    int my_col = (mv_col[1]<<y_shift)/2;
1234
                    int mx = (scale * mv_col[0] + 128) >> 8;
1235
                    int my = (scale * my_col + 128) >> 8;
1236
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1237
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1238
                }
1239
            }
1240
            return;
1241
        }
1242

    
1243
        /* one-to-one mv scaling */
1244

    
1245
        if(IS_16X16(*mb_type)){
1246
            int ref, mv0, mv1;
1247

    
1248
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1249
            if(IS_INTRA(mb_type_col[0])){
1250
                ref=mv0=mv1=0;
1251
            }else{
1252
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
1253
                                                : map_col_to_list0[1][l1ref1[0] + ref_offset];
1254
                const int scale = dist_scale_factor[ref0];
1255
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1256
                int mv_l0[2];
1257
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1258
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1259
                ref= ref0;
1260
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
1261
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1262
            }
1263
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1264
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1265
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1266
        }else{
1267
            for(i8=0; i8<4; i8++){
1268
                const int x8 = i8&1;
1269
                const int y8 = i8>>1;
1270
                int ref0, scale;
1271
                const int16_t (*l1mv)[2]= l1mv0;
1272

    
1273
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1274
                    continue;
1275
                h->sub_mb_type[i8] = sub_mb_type;
1276
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1277
                if(IS_INTRA(mb_type_col[0])){
1278
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1279
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1280
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1281
                    continue;
1282
                }
1283

    
1284
                ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset;
1285
                if(ref0 >= 0)
1286
                    ref0 = map_col_to_list0[0][ref0];
1287
                else{
1288
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1289
                    l1mv= l1mv1;
1290
                }
1291
                scale = dist_scale_factor[ref0];
1292

    
1293
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1294
                if(IS_SUB_8X8(sub_mb_type)){
1295
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1296
                    int mx = (scale * mv_col[0] + 128) >> 8;
1297
                    int my = (scale * mv_col[1] + 128) >> 8;
1298
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1299
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1300
                }else
1301
                for(i4=0; i4<4; i4++){
1302
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1303
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1304
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1305
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1306
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1307
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1308
                }
1309
            }
1310
        }
1311
    }
1312
}
1313

    
1314
static inline void write_back_motion(H264Context *h, int mb_type){
1315
    MpegEncContext * const s = &h->s;
1316
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1317
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1318
    int list;
1319

    
1320
    if(!USES_LIST(mb_type, 0))
1321
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1322

    
1323
    for(list=0; list<h->list_count; list++){
1324
        int y;
1325
        if(!USES_LIST(mb_type, list))
1326
            continue;
1327

    
1328
        for(y=0; y<4; y++){
1329
            *(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];
1330
            *(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];
1331
        }
1332
        if( h->pps.cabac ) {
1333
            if(IS_SKIP(mb_type))
1334
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1335
            else
1336
            for(y=0; y<4; y++){
1337
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1338
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1339
            }
1340
        }
1341

    
1342
        {
1343
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1344
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1345
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1346
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1347
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1348
        }
1349
    }
1350

    
1351
    if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
1352
        if(IS_8X8(mb_type)){
1353
            uint8_t *direct_table = &h->direct_table[b8_xy];
1354
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1355
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1356
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1357
        }
1358
    }
1359
}
1360

    
1361
/**
1362
 * Decodes a network abstraction layer unit.
1363
 * @param consumed is the number of bytes used as input
1364
 * @param length is the length of the array
1365
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1366
 * @returns decoded bytes, might be src+1 if no escapes
1367
 */
1368
static const uint8_t *decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1369
    int i, si, di;
1370
    uint8_t *dst;
1371
    int bufidx;
1372

    
1373
//    src[0]&0x80;                //forbidden bit
1374
    h->nal_ref_idc= src[0]>>5;
1375
    h->nal_unit_type= src[0]&0x1F;
1376

    
1377
    src++; length--;
1378
#if 0
1379
    for(i=0; i<length; i++)
1380
        printf("%2X ", src[i]);
1381
#endif
1382
    for(i=0; i+1<length; i+=2){
1383
        if(src[i]) continue;
1384
        if(i>0 && src[i-1]==0) i--;
1385
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1386
            if(src[i+2]!=3){
1387
                /* startcode, so we must be past the end */
1388
                length=i;
1389
            }
1390
            break;
1391
        }
1392
    }
1393

    
1394
    if(i>=length-1){ //no escaped 0
1395
        *dst_length= length;
1396
        *consumed= length+1; //+1 for the header
1397
        return src;
1398
    }
1399

    
1400
    bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1401
    h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
1402
    dst= h->rbsp_buffer[bufidx];
1403

    
1404
    if (dst == NULL){
1405
        return NULL;
1406
    }
1407

    
1408
//printf("decoding esc\n");
1409
    si=di=0;
1410
    while(si<length){
1411
        //remove escapes (very rare 1:2^22)
1412
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1413
            if(src[si+2]==3){ //escape
1414
                dst[di++]= 0;
1415
                dst[di++]= 0;
1416
                si+=3;
1417
                continue;
1418
            }else //next start code
1419
                break;
1420
        }
1421

    
1422
        dst[di++]= src[si++];
1423
    }
1424

    
1425
    memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
1426

    
1427
    *dst_length= di;
1428
    *consumed= si + 1;//+1 for the header
1429
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1430
    return dst;
1431
}
1432

    
1433
/**
1434
 * identifies the exact end of the bitstream
1435
 * @return the length of the trailing, or 0 if damaged
1436
 */
1437
static int decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1438
    int v= *src;
1439
    int r;
1440

    
1441
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1442

    
1443
    for(r=1; r<9; r++){
1444
        if(v&1) return r;
1445
        v>>=1;
1446
    }
1447
    return 0;
1448
}
1449

    
1450
/**
1451
 * IDCT transforms the 16 dc values and dequantizes them.
1452
 * @param qp quantization parameter
1453
 */
1454
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1455
#define stride 16
1456
    int i;
1457
    int temp[16]; //FIXME check if this is a good idea
1458
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1459
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1460

    
1461
//memset(block, 64, 2*256);
1462
//return;
1463
    for(i=0; i<4; i++){
1464
        const int offset= y_offset[i];
1465
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1466
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1467
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1468
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1469

    
1470
        temp[4*i+0]= z0+z3;
1471
        temp[4*i+1]= z1+z2;
1472
        temp[4*i+2]= z1-z2;
1473
        temp[4*i+3]= z0-z3;
1474
    }
1475

    
1476
    for(i=0; i<4; i++){
1477
        const int offset= x_offset[i];
1478
        const int z0= temp[4*0+i] + temp[4*2+i];
1479
        const int z1= temp[4*0+i] - temp[4*2+i];
1480
        const int z2= temp[4*1+i] - temp[4*3+i];
1481
        const int z3= temp[4*1+i] + temp[4*3+i];
1482

    
1483
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
1484
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1485
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1486
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1487
    }
1488
}
1489

    
1490
#if 0
1491
/**
1492
 * DCT transforms the 16 dc values.
1493
 * @param qp quantization parameter ??? FIXME
1494
 */
1495
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1496
//    const int qmul= dequant_coeff[qp][0];
1497
    int i;
1498
    int temp[16]; //FIXME check if this is a good idea
1499
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1500
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1501

1502
    for(i=0; i<4; i++){
1503
        const int offset= y_offset[i];
1504
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1505
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1506
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1507
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1508

1509
        temp[4*i+0]= z0+z3;
1510
        temp[4*i+1]= z1+z2;
1511
        temp[4*i+2]= z1-z2;
1512
        temp[4*i+3]= z0-z3;
1513
    }
1514

1515
    for(i=0; i<4; i++){
1516
        const int offset= x_offset[i];
1517
        const int z0= temp[4*0+i] + temp[4*2+i];
1518
        const int z1= temp[4*0+i] - temp[4*2+i];
1519
        const int z2= temp[4*1+i] - temp[4*3+i];
1520
        const int z3= temp[4*1+i] + temp[4*3+i];
1521

1522
        block[stride*0 +offset]= (z0 + z3)>>1;
1523
        block[stride*2 +offset]= (z1 + z2)>>1;
1524
        block[stride*8 +offset]= (z1 - z2)>>1;
1525
        block[stride*10+offset]= (z0 - z3)>>1;
1526
    }
1527
}
1528
#endif
1529

    
1530
#undef xStride
1531
#undef stride
1532

    
1533
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1534
    const int stride= 16*2;
1535
    const int xStride= 16;
1536
    int a,b,c,d,e;
1537

    
1538
    a= block[stride*0 + xStride*0];
1539
    b= block[stride*0 + xStride*1];
1540
    c= block[stride*1 + xStride*0];
1541
    d= block[stride*1 + xStride*1];
1542

    
1543
    e= a-b;
1544
    a= a+b;
1545
    b= c-d;
1546
    c= c+d;
1547

    
1548
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1549
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1550
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1551
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1552
}
1553

    
1554
#if 0
1555
static void chroma_dc_dct_c(DCTELEM *block){
1556
    const int stride= 16*2;
1557
    const int xStride= 16;
1558
    int a,b,c,d,e;
1559

1560
    a= block[stride*0 + xStride*0];
1561
    b= block[stride*0 + xStride*1];
1562
    c= block[stride*1 + xStride*0];
1563
    d= block[stride*1 + xStride*1];
1564

1565
    e= a-b;
1566
    a= a+b;
1567
    b= c-d;
1568
    c= c+d;
1569

1570
    block[stride*0 + xStride*0]= (a+c);
1571
    block[stride*0 + xStride*1]= (e+b);
1572
    block[stride*1 + xStride*0]= (a-c);
1573
    block[stride*1 + xStride*1]= (e-b);
1574
}
1575
#endif
1576

    
1577
/**
1578
 * gets the chroma qp.
1579
 */
1580
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1581
    return h->pps.chroma_qp_table[t][qscale];
1582
}
1583

    
1584
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1585
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1586
                           int src_x_offset, int src_y_offset,
1587
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1588
    MpegEncContext * const s = &h->s;
1589
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1590
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1591
    const int luma_xy= (mx&3) + ((my&3)<<2);
1592
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1593
    uint8_t * src_cb, * src_cr;
1594
    int extra_width= h->emu_edge_width;
1595
    int extra_height= h->emu_edge_height;
1596
    int emu=0;
1597
    const int full_mx= mx>>2;
1598
    const int full_my= my>>2;
1599
    const int pic_width  = 16*s->mb_width;
1600
    const int pic_height = 16*s->mb_height >> MB_FIELD;
1601

    
1602
    if(mx&7) extra_width -= 3;
1603
    if(my&7) extra_height -= 3;
1604

    
1605
    if(   full_mx < 0-extra_width
1606
       || full_my < 0-extra_height
1607
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
1608
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
1609
        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);
1610
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1611
        emu=1;
1612
    }
1613

    
1614
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1615
    if(!square){
1616
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1617
    }
1618

    
1619
    if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1620

    
1621
    if(MB_FIELD){
1622
        // chroma offset when predicting from a field of opposite parity
1623
        my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1624
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1625
    }
1626
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1627
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1628

    
1629
    if(emu){
1630
        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);
1631
            src_cb= s->edge_emu_buffer;
1632
    }
1633
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1634

    
1635
    if(emu){
1636
        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);
1637
            src_cr= s->edge_emu_buffer;
1638
    }
1639
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1640
}
1641

    
1642
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1643
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1644
                           int x_offset, int y_offset,
1645
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1646
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1647
                           int list0, int list1){
1648
    MpegEncContext * const s = &h->s;
1649
    qpel_mc_func *qpix_op=  qpix_put;
1650
    h264_chroma_mc_func chroma_op= chroma_put;
1651

    
1652
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1653
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1654
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1655
    x_offset += 8*s->mb_x;
1656
    y_offset += 8*(s->mb_y >> MB_FIELD);
1657

    
1658
    if(list0){
1659
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1660
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1661
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1662
                           qpix_op, chroma_op);
1663

    
1664
        qpix_op=  qpix_avg;
1665
        chroma_op= chroma_avg;
1666
    }
1667

    
1668
    if(list1){
1669
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1670
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1671
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1672
                           qpix_op, chroma_op);
1673
    }
1674
}
1675

    
1676
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1677
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1678
                           int x_offset, int y_offset,
1679
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1680
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1681
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1682
                           int list0, int list1){
1683
    MpegEncContext * const s = &h->s;
1684

    
1685
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1686
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1687
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1688
    x_offset += 8*s->mb_x;
1689
    y_offset += 8*(s->mb_y >> MB_FIELD);
1690

    
1691
    if(list0 && list1){
1692
        /* don't optimize for luma-only case, since B-frames usually
1693
         * use implicit weights => chroma too. */
1694
        uint8_t *tmp_cb = s->obmc_scratchpad;
1695
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1696
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1697
        int refn0 = h->ref_cache[0][ scan8[n] ];
1698
        int refn1 = h->ref_cache[1][ scan8[n] ];
1699

    
1700
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1701
                    dest_y, dest_cb, dest_cr,
1702
                    x_offset, y_offset, qpix_put, chroma_put);
1703
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1704
                    tmp_y, tmp_cb, tmp_cr,
1705
                    x_offset, y_offset, qpix_put, chroma_put);
1706

    
1707
        if(h->use_weight == 2){
1708
            int weight0 = h->implicit_weight[refn0][refn1];
1709
            int weight1 = 64 - weight0;
1710
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
1711
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1712
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1713
        }else{
1714
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1715
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1716
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1717
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1718
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1719
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1720
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1721
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1722
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1723
        }
1724
    }else{
1725
        int list = list1 ? 1 : 0;
1726
        int refn = h->ref_cache[list][ scan8[n] ];
1727
        Picture *ref= &h->ref_list[list][refn];
1728
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1729
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
1730
                    qpix_put, chroma_put);
1731

    
1732
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1733
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
1734
        if(h->use_weight_chroma){
1735
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1736
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1737
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1738
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1739
        }
1740
    }
1741
}
1742

    
1743
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1744
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1745
                           int x_offset, int y_offset,
1746
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1747
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1748
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1749
                           int list0, int list1){
1750
    if((h->use_weight==2 && list0 && list1
1751
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1752
       || h->use_weight==1)
1753
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1754
                         x_offset, y_offset, qpix_put, chroma_put,
1755
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1756
    else
1757
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1758
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1759
}
1760

    
1761
static inline void prefetch_motion(H264Context *h, int list){
1762
    /* fetch pixels for estimated mv 4 macroblocks ahead
1763
     * optimized for 64byte cache lines */
1764
    MpegEncContext * const s = &h->s;
1765
    const int refn = h->ref_cache[list][scan8[0]];
1766
    if(refn >= 0){
1767
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1768
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1769
        uint8_t **src= h->ref_list[list][refn].data;
1770
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1771
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
1772
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1773
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1774
    }
1775
}
1776

    
1777
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1778
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1779
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1780
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1781
    MpegEncContext * const s = &h->s;
1782
    const int mb_xy= h->mb_xy;
1783
    const int mb_type= s->current_picture.mb_type[mb_xy];
1784

    
1785
    assert(IS_INTER(mb_type));
1786

    
1787
    prefetch_motion(h, 0);
1788

    
1789
    if(IS_16X16(mb_type)){
1790
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1791
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1792
                &weight_op[0], &weight_avg[0],
1793
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1794
    }else if(IS_16X8(mb_type)){
1795
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1796
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1797
                &weight_op[1], &weight_avg[1],
1798
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1799
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1800
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1801
                &weight_op[1], &weight_avg[1],
1802
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1803
    }else if(IS_8X16(mb_type)){
1804
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1805
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1806
                &weight_op[2], &weight_avg[2],
1807
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1808
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1809
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1810
                &weight_op[2], &weight_avg[2],
1811
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1812
    }else{
1813
        int i;
1814

    
1815
        assert(IS_8X8(mb_type));
1816

    
1817
        for(i=0; i<4; i++){
1818
            const int sub_mb_type= h->sub_mb_type[i];
1819
            const int n= 4*i;
1820
            int x_offset= (i&1)<<2;
1821
            int y_offset= (i&2)<<1;
1822

    
1823
            if(IS_SUB_8X8(sub_mb_type)){
1824
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1825
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1826
                    &weight_op[3], &weight_avg[3],
1827
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1828
            }else if(IS_SUB_8X4(sub_mb_type)){
1829
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1830
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1831
                    &weight_op[4], &weight_avg[4],
1832
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1833
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1834
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1835
                    &weight_op[4], &weight_avg[4],
1836
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1837
            }else if(IS_SUB_4X8(sub_mb_type)){
1838
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1839
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1840
                    &weight_op[5], &weight_avg[5],
1841
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1842
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1843
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1844
                    &weight_op[5], &weight_avg[5],
1845
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1846
            }else{
1847
                int j;
1848
                assert(IS_SUB_4X4(sub_mb_type));
1849
                for(j=0; j<4; j++){
1850
                    int sub_x_offset= x_offset + 2*(j&1);
1851
                    int sub_y_offset= y_offset +   (j&2);
1852
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1853
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1854
                        &weight_op[6], &weight_avg[6],
1855
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1856
                }
1857
            }
1858
        }
1859
    }
1860

    
1861
    prefetch_motion(h, 1);
1862
}
1863

    
1864
static av_cold void decode_init_vlc(void){
1865
    static int done = 0;
1866

    
1867
    if (!done) {
1868
        int i;
1869
        int offset;
1870
        done = 1;
1871

    
1872
        chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
1873
        chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
1874
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1875
                 &chroma_dc_coeff_token_len [0], 1, 1,
1876
                 &chroma_dc_coeff_token_bits[0], 1, 1,
1877
                 INIT_VLC_USE_NEW_STATIC);
1878

    
1879
        offset = 0;
1880
        for(i=0; i<4; i++){
1881
            coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
1882
            coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
1883
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1884
                     &coeff_token_len [i][0], 1, 1,
1885
                     &coeff_token_bits[i][0], 1, 1,
1886
                     INIT_VLC_USE_NEW_STATIC);
1887
            offset += coeff_token_vlc_tables_size[i];
1888
        }
1889
        /*
1890
         * This is a one time safety check to make sure that
1891
         * the packed static coeff_token_vlc table sizes
1892
         * were initialized correctly.
1893
         */
1894
        assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
1895

    
1896
        for(i=0; i<3; i++){
1897
            chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
1898
            chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
1899
            init_vlc(&chroma_dc_total_zeros_vlc[i],
1900
                     CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1901
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
1902
                     &chroma_dc_total_zeros_bits[i][0], 1, 1,
1903
                     INIT_VLC_USE_NEW_STATIC);
1904
        }
1905
        for(i=0; i<15; i++){
1906
            total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
1907
            total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
1908
            init_vlc(&total_zeros_vlc[i],
1909
                     TOTAL_ZEROS_VLC_BITS, 16,
1910
                     &total_zeros_len [i][0], 1, 1,
1911
                     &total_zeros_bits[i][0], 1, 1,
1912
                     INIT_VLC_USE_NEW_STATIC);
1913
        }
1914

    
1915
        for(i=0; i<6; i++){
1916
            run_vlc[i].table = run_vlc_tables[i];
1917
            run_vlc[i].table_allocated = run_vlc_tables_size;
1918
            init_vlc(&run_vlc[i],
1919
                     RUN_VLC_BITS, 7,
1920
                     &run_len [i][0], 1, 1,
1921
                     &run_bits[i][0], 1, 1,
1922
                     INIT_VLC_USE_NEW_STATIC);
1923
        }
1924
        run7_vlc.table = run7_vlc_table,
1925
        run7_vlc.table_allocated = run7_vlc_table_size;
1926
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1927
                 &run_len [6][0], 1, 1,
1928
                 &run_bits[6][0], 1, 1,
1929
                 INIT_VLC_USE_NEW_STATIC);
1930
    }
1931
}
1932

    
1933
static void free_tables(H264Context *h){
1934
    int i;
1935
    H264Context *hx;
1936
    av_freep(&h->intra4x4_pred_mode);
1937
    av_freep(&h->chroma_pred_mode_table);
1938
    av_freep(&h->cbp_table);
1939
    av_freep(&h->mvd_table[0]);
1940
    av_freep(&h->mvd_table[1]);
1941
    av_freep(&h->direct_table);
1942
    av_freep(&h->non_zero_count);
1943
    av_freep(&h->slice_table_base);
1944
    h->slice_table= NULL;
1945

    
1946
    av_freep(&h->mb2b_xy);
1947
    av_freep(&h->mb2b8_xy);
1948

    
1949
    for(i = 0; i < h->s.avctx->thread_count; i++) {
1950
        hx = h->thread_context[i];
1951
        if(!hx) continue;
1952
        av_freep(&hx->top_borders[1]);
1953
        av_freep(&hx->top_borders[0]);
1954
        av_freep(&hx->s.obmc_scratchpad);
1955
    }
1956
}
1957

    
1958
static void init_dequant8_coeff_table(H264Context *h){
1959
    int i,q,x;
1960
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
1961
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
1962
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
1963

    
1964
    for(i=0; i<2; i++ ){
1965
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
1966
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
1967
            break;
1968
        }
1969

    
1970
        for(q=0; q<52; q++){
1971
            int shift = div6[q];
1972
            int idx = rem6[q];
1973
            for(x=0; x<64; x++)
1974
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
1975
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
1976
                    h->pps.scaling_matrix8[i][x]) << shift;
1977
        }
1978
    }
1979
}
1980

    
1981
static void init_dequant4_coeff_table(H264Context *h){
1982
    int i,j,q,x;
1983
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
1984
    for(i=0; i<6; i++ ){
1985
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
1986
        for(j=0; j<i; j++){
1987
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
1988
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
1989
                break;
1990
            }
1991
        }
1992
        if(j<i)
1993
            continue;
1994

    
1995
        for(q=0; q<52; q++){
1996
            int shift = div6[q] + 2;
1997
            int idx = rem6[q];
1998
            for(x=0; x<16; x++)
1999
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2000
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2001
                    h->pps.scaling_matrix4[i][x]) << shift;
2002
        }
2003
    }
2004
}
2005

    
2006
static void init_dequant_tables(H264Context *h){
2007
    int i,x;
2008
    init_dequant4_coeff_table(h);
2009
    if(h->pps.transform_8x8_mode)
2010
        init_dequant8_coeff_table(h);
2011
    if(h->sps.transform_bypass){
2012
        for(i=0; i<6; i++)
2013
            for(x=0; x<16; x++)
2014
                h->dequant4_coeff[i][0][x] = 1<<6;
2015
        if(h->pps.transform_8x8_mode)
2016
            for(i=0; i<2; i++)
2017
                for(x=0; x<64; x++)
2018
                    h->dequant8_coeff[i][0][x] = 1<<6;
2019
    }
2020
}
2021

    
2022

    
2023
/**
2024
 * allocates tables.
2025
 * needs width/height
2026
 */
2027
static int alloc_tables(H264Context *h){
2028
    MpegEncContext * const s = &h->s;
2029
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2030
    int x,y;
2031

    
2032
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2033

    
2034
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2035
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base))
2036
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2037

    
2038
    CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2039
    CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2040
    CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2041
    CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2042

    
2043
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
2044
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2045

    
2046
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2047
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2048
    for(y=0; y<s->mb_height; y++){
2049
        for(x=0; x<s->mb_width; x++){
2050
            const int mb_xy= x + y*s->mb_stride;
2051
            const int b_xy = 4*x + 4*y*h->b_stride;
2052
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2053

    
2054
            h->mb2b_xy [mb_xy]= b_xy;
2055
            h->mb2b8_xy[mb_xy]= b8_xy;
2056
        }
2057
    }
2058

    
2059
    s->obmc_scratchpad = NULL;
2060

    
2061
    if(!h->dequant4_coeff[0])
2062
        init_dequant_tables(h);
2063

    
2064
    return 0;
2065
fail:
2066
    free_tables(h);
2067
    return -1;
2068
}
2069

    
2070
/**
2071
 * Mimic alloc_tables(), but for every context thread.
2072
 */
2073
static void clone_tables(H264Context *dst, H264Context *src){
2074
    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
2075
    dst->non_zero_count           = src->non_zero_count;
2076
    dst->slice_table              = src->slice_table;
2077
    dst->cbp_table                = src->cbp_table;
2078
    dst->mb2b_xy                  = src->mb2b_xy;
2079
    dst->mb2b8_xy                 = src->mb2b8_xy;
2080
    dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
2081
    dst->mvd_table[0]             = src->mvd_table[0];
2082
    dst->mvd_table[1]             = src->mvd_table[1];
2083
    dst->direct_table             = src->direct_table;
2084

    
2085
    dst->s.obmc_scratchpad = NULL;
2086
    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2087
}
2088

    
2089
/**
2090
 * Init context
2091
 * Allocate buffers which are not shared amongst multiple threads.
2092
 */
2093
static int context_init(H264Context *h){
2094
    CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2095
    CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2096

    
2097
    return 0;
2098
fail:
2099
    return -1; // free_tables will clean up for us
2100
}
2101

    
2102
static av_cold void common_init(H264Context *h){
2103
    MpegEncContext * const s = &h->s;
2104

    
2105
    s->width = s->avctx->width;
2106
    s->height = s->avctx->height;
2107
    s->codec_id= s->avctx->codec->id;
2108

    
2109
    ff_h264_pred_init(&h->hpc, s->codec_id);
2110

    
2111
    h->dequant_coeff_pps= -1;
2112
    s->unrestricted_mv=1;
2113
    s->decode=1; //FIXME
2114

    
2115
    dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
2116

    
2117
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2118
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2119
}
2120

    
2121
static av_cold int decode_init(AVCodecContext *avctx){
2122
    H264Context *h= avctx->priv_data;
2123
    MpegEncContext * const s = &h->s;
2124

    
2125
    MPV_decode_defaults(s);
2126

    
2127
    s->avctx = avctx;
2128
    common_init(h);
2129

    
2130
    s->out_format = FMT_H264;
2131
    s->workaround_bugs= avctx->workaround_bugs;
2132

    
2133
    // set defaults
2134
//    s->decode_mb= ff_h263_decode_mb;
2135
    s->quarter_sample = 1;
2136
    s->low_delay= 1;
2137

    
2138
    if(avctx->codec_id == CODEC_ID_SVQ3)
2139
        avctx->pix_fmt= PIX_FMT_YUVJ420P;
2140
    else
2141
        avctx->pix_fmt= PIX_FMT_YUV420P;
2142

    
2143
    decode_init_vlc();
2144

    
2145
    if(avctx->extradata_size > 0 && avctx->extradata &&
2146
       *(char *)avctx->extradata == 1){
2147
        h->is_avc = 1;
2148
        h->got_avcC = 0;
2149
    } else {
2150
        h->is_avc = 0;
2151
    }
2152

    
2153
    h->thread_context[0] = h;
2154
    h->outputed_poc = INT_MIN;
2155
    h->prev_poc_msb= 1<<16;
2156
    return 0;
2157
}
2158

    
2159
static int frame_start(H264Context *h){
2160
    MpegEncContext * const s = &h->s;
2161
    int i;
2162

    
2163
    if(MPV_frame_start(s, s->avctx) < 0)
2164
        return -1;
2165
    ff_er_frame_start(s);
2166
    /*
2167
     * MPV_frame_start uses pict_type to derive key_frame.
2168
     * This is incorrect for H.264; IDR markings must be used.
2169
     * Zero here; IDR markings per slice in frame or fields are ORed in later.
2170
     * See decode_nal_units().
2171
     */
2172
    s->current_picture_ptr->key_frame= 0;
2173

    
2174
    assert(s->linesize && s->uvlinesize);
2175

    
2176
    for(i=0; i<16; i++){
2177
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2178
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2179
    }
2180
    for(i=0; i<4; i++){
2181
        h->block_offset[16+i]=
2182
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2183
        h->block_offset[24+16+i]=
2184
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2185
    }
2186

    
2187
    /* can't be in alloc_tables because linesize isn't known there.
2188
     * FIXME: redo bipred weight to not require extra buffer? */
2189
    for(i = 0; i < s->avctx->thread_count; i++)
2190
        if(!h->thread_context[i]->s.obmc_scratchpad)
2191
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2192

    
2193
    /* some macroblocks will be accessed before they're available */
2194
    if(FRAME_MBAFF || s->avctx->thread_count > 1)
2195
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
2196

    
2197
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2198

    
2199
    // We mark the current picture as non-reference after allocating it, so
2200
    // that if we break out due to an error it can be released automatically
2201
    // in the next MPV_frame_start().
2202
    // SVQ3 as well as most other codecs have only last/next/current and thus
2203
    // get released even with set reference, besides SVQ3 and others do not
2204
    // mark frames as reference later "naturally".
2205
    if(s->codec_id != CODEC_ID_SVQ3)
2206
        s->current_picture_ptr->reference= 0;
2207

    
2208
    s->current_picture_ptr->field_poc[0]=
2209
    s->current_picture_ptr->field_poc[1]= INT_MAX;
2210
    assert(s->current_picture_ptr->long_ref==0);
2211

    
2212
    return 0;
2213
}
2214

    
2215
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){
2216
    MpegEncContext * const s = &h->s;
2217
    int i;
2218
    int step    = 1;
2219
    int offset  = 1;
2220
    int uvoffset= 1;
2221
    int top_idx = 1;
2222
    int skiplast= 0;
2223

    
2224
    src_y  -=   linesize;
2225
    src_cb -= uvlinesize;
2226
    src_cr -= uvlinesize;
2227

    
2228
    if(!simple && FRAME_MBAFF){
2229
        if(s->mb_y&1){
2230
            offset  = MB_MBAFF ? 1 : 17;
2231
            uvoffset= MB_MBAFF ? 1 : 9;
2232
            if(!MB_MBAFF){
2233
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y +  15*linesize);
2234
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
2235
                if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2236
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
2237
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
2238
                }
2239
            }
2240
        }else{
2241
            if(!MB_MBAFF){
2242
                h->left_border[0]= h->top_borders[0][s->mb_x][15];
2243
                if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2244
                    h->left_border[34   ]= h->top_borders[0][s->mb_x][16+7  ];
2245
                    h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
2246
                }
2247
                skiplast= 1;
2248
            }
2249
            offset  =
2250
            uvoffset=
2251
            top_idx = MB_MBAFF ? 0 : 1;
2252
        }
2253
        step= MB_MBAFF ? 2 : 1;
2254
    }
2255

    
2256
    // There are two lines saved, the line above the the top macroblock of a pair,
2257
    // and the line above the bottom macroblock
2258
    h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
2259
    for(i=1; i<17 - skiplast; i++){
2260
        h->left_border[offset+i*step]= src_y[15+i*  linesize];
2261
    }
2262

    
2263
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2264
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2265

    
2266
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2267
        h->left_border[uvoffset+34   ]= h->top_borders[top_idx][s->mb_x][16+7];
2268
        h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
2269
        for(i=1; i<9 - skiplast; i++){
2270
            h->left_border[uvoffset+34   +i*step]= src_cb[7+i*uvlinesize];
2271
            h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
2272
        }
2273
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2274
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2275
    }
2276
}
2277

    
2278
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){
2279
    MpegEncContext * const s = &h->s;
2280
    int temp8, i;
2281
    uint64_t temp64;
2282
    int deblock_left;
2283
    int deblock_top;
2284
    int mb_xy;
2285
    int step    = 1;
2286
    int offset  = 1;
2287
    int uvoffset= 1;
2288
    int top_idx = 1;
2289

    
2290
    if(!simple && FRAME_MBAFF){
2291
        if(s->mb_y&1){
2292
            offset  = MB_MBAFF ? 1 : 17;
2293
            uvoffset= MB_MBAFF ? 1 : 9;
2294
        }else{
2295
            offset  =
2296
            uvoffset=
2297
            top_idx = MB_MBAFF ? 0 : 1;
2298
        }
2299
        step= MB_MBAFF ? 2 : 1;
2300
    }
2301

    
2302
    if(h->deblocking_filter == 2) {
2303
        mb_xy = h->mb_xy;
2304
        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2305
        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2306
    } else {
2307
        deblock_left = (s->mb_x > 0);
2308
        deblock_top =  (s->mb_y > !!MB_FIELD);
2309
    }
2310

    
2311
    src_y  -=   linesize + 1;
2312
    src_cb -= uvlinesize + 1;
2313
    src_cr -= uvlinesize + 1;
2314

    
2315
#define XCHG(a,b,t,xchg)\
2316
t= a;\
2317
if(xchg)\
2318
    a= b;\
2319
b= t;
2320

    
2321
    if(deblock_left){
2322
        for(i = !deblock_top; i<16; i++){
2323
            XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, xchg);
2324
        }
2325
        XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, 1);
2326
    }
2327

    
2328
    if(deblock_top){
2329
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2330
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2331
        if(s->mb_x+1 < s->mb_width){
2332
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2333
        }
2334
    }
2335

    
2336
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2337
        if(deblock_left){
2338
            for(i = !deblock_top; i<8; i++){
2339
                XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, xchg);
2340
                XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
2341
            }
2342
            XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, 1);
2343
            XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
2344
        }
2345
        if(deblock_top){
2346
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2347
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2348
        }
2349
    }
2350
}
2351

    
2352
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2353
    MpegEncContext * const s = &h->s;
2354
    const int mb_x= s->mb_x;
2355
    const int mb_y= s->mb_y;
2356
    const int mb_xy= h->mb_xy;
2357
    const int mb_type= s->current_picture.mb_type[mb_xy];
2358
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2359
    int linesize, uvlinesize /*dct_offset*/;
2360
    int i;
2361
    int *block_offset = &h->block_offset[0];
2362
    const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2363
    const int is_h264 = simple || s->codec_id == CODEC_ID_H264;
2364
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2365
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2366

    
2367
    dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2368
    dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2369
    dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2370

    
2371
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2372
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2373

    
2374
    if (!simple && MB_FIELD) {
2375
        linesize   = h->mb_linesize   = s->linesize * 2;
2376
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2377
        block_offset = &h->block_offset[24];
2378
        if(mb_y&1){ //FIXME move out of this function?
2379
            dest_y -= s->linesize*15;
2380
            dest_cb-= s->uvlinesize*7;
2381
            dest_cr-= s->uvlinesize*7;
2382
        }
2383
        if(FRAME_MBAFF) {
2384
            int list;
2385
            for(list=0; list<h->list_count; list++){
2386
                if(!USES_LIST(mb_type, list))
2387
                    continue;
2388
                if(IS_16X16(mb_type)){
2389
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
2390
                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2391
                }else{
2392
                    for(i=0; i<16; i+=4){
2393
                        int ref = h->ref_cache[list][scan8[i]];
2394
                        if(ref >= 0)
2395
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2396
                    }
2397
                }
2398
            }
2399
        }
2400
    } else {
2401
        linesize   = h->mb_linesize   = s->linesize;
2402
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2403
//        dct_offset = s->linesize * 16;
2404
    }
2405

    
2406
    if (!simple && IS_INTRA_PCM(mb_type)) {
2407
        for (i=0; i<16; i++) {
2408
            memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
2409
        }
2410
        for (i=0; i<8; i++) {
2411
            memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
2412
            memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
2413
        }
2414
    } else {
2415
        if(IS_INTRA(mb_type)){
2416
            if(h->deblocking_filter)
2417
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2418

    
2419
            if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2420
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2421
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2422
            }
2423

    
2424
            if(IS_INTRA4x4(mb_type)){
2425
                if(simple || !s->encoding){
2426
                    if(IS_8x8DCT(mb_type)){
2427
                        if(transform_bypass){
2428
                            idct_dc_add =
2429
                            idct_add    = s->dsp.add_pixels8;
2430
                        }else{
2431
                            idct_dc_add = s->dsp.h264_idct8_dc_add;
2432
                            idct_add    = s->dsp.h264_idct8_add;
2433
                        }
2434
                        for(i=0; i<16; i+=4){
2435
                            uint8_t * const ptr= dest_y + block_offset[i];
2436
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2437
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2438
                                h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
2439
                            }else{
2440
                                const int nnz = h->non_zero_count_cache[ scan8[i] ];
2441
                                h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2442
                                                            (h->topright_samples_available<<i)&0x4000, linesize);
2443
                                if(nnz){
2444
                                    if(nnz == 1 && h->mb[i*16])
2445
                                        idct_dc_add(ptr, h->mb + i*16, linesize);
2446
                                    else
2447
                                        idct_add   (ptr, h->mb + i*16, linesize);
2448
                                }
2449
                            }
2450
                        }
2451
                    }else{
2452
                        if(transform_bypass){
2453
                            idct_dc_add =
2454
                            idct_add    = s->dsp.add_pixels4;
2455
                        }else{
2456
                            idct_dc_add = s->dsp.h264_idct_dc_add;
2457
                            idct_add    = s->dsp.h264_idct_add;
2458
                        }
2459
                        for(i=0; i<16; i++){
2460
                            uint8_t * const ptr= dest_y + block_offset[i];
2461
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2462

    
2463
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2464
                                h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
2465
                            }else{
2466
                                uint8_t *topright;
2467
                                int nnz, tr;
2468
                                if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2469
                                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2470
                                    assert(mb_y || linesize <= block_offset[i]);
2471
                                    if(!topright_avail){
2472
                                        tr= ptr[3 - linesize]*0x01010101;
2473
                                        topright= (uint8_t*) &tr;
2474
                                    }else
2475
                                        topright= ptr + 4 - linesize;
2476
                                }else
2477
                                    topright= NULL;
2478

    
2479
                                h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2480
                                nnz = h->non_zero_count_cache[ scan8[i] ];
2481
                                if(nnz){
2482
                                    if(is_h264){
2483
                                        if(nnz == 1 && h->mb[i*16])
2484
                                            idct_dc_add(ptr, h->mb + i*16, linesize);
2485
                                        else
2486
                                            idct_add   (ptr, h->mb + i*16, linesize);
2487
                                    }else
2488
                                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2489
                                }
2490
                            }
2491
                        }
2492
                    }
2493
                }
2494
            }else{
2495
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2496
                if(is_h264){
2497
                    if(!transform_bypass)
2498
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
2499
                }else
2500
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2501
            }
2502
            if(h->deblocking_filter)
2503
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2504
        }else if(is_h264){
2505
            hl_motion(h, dest_y, dest_cb, dest_cr,
2506
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2507
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2508
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2509
        }
2510

    
2511

    
2512
        if(!IS_INTRA4x4(mb_type)){
2513
            if(is_h264){
2514
                if(IS_INTRA16x16(mb_type)){
2515
                    if(transform_bypass){
2516
                        if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
2517
                            h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
2518
                        }else{
2519
                            for(i=0; i<16; i++){
2520
                                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2521
                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
2522
                            }
2523
                        }
2524
                    }else{
2525
                         s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2526
                    }
2527
                }else if(h->cbp&15){
2528
                    if(transform_bypass){
2529
                        const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2530
                        idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2531
                        for(i=0; i<16; i+=di){
2532
                            if(h->non_zero_count_cache[ scan8[i] ]){
2533
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2534
                            }
2535
                        }
2536
                    }else{
2537
                        if(IS_8x8DCT(mb_type)){
2538
                            s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2539
                        }else{
2540
                            s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2541
                        }
2542
                    }
2543
                }
2544
            }else{
2545
                for(i=0; i<16; i++){
2546
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2547
                        uint8_t * const ptr= dest_y + block_offset[i];
2548
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2549
                    }
2550
                }
2551
            }
2552
        }
2553

    
2554
        if((simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2555
            uint8_t *dest[2] = {dest_cb, dest_cr};
2556
            if(transform_bypass){
2557
                if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2558
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
2559
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
2560
                }else{
2561
                    idct_add = s->dsp.add_pixels4;
2562
                    for(i=16; i<16+8; i++){
2563
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2564
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2565
                    }
2566
                }
2567
            }else{
2568
                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]);
2569
                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]);
2570
                if(is_h264){
2571
                    idct_add = s->dsp.h264_idct_add;
2572
                    idct_dc_add = s->dsp.h264_idct_dc_add;
2573
                    for(i=16; i<16+8; i++){
2574
                        if(h->non_zero_count_cache[ scan8[i] ])
2575
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2576
                        else if(h->mb[i*16])
2577
                            idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2578
                    }
2579
                }else{
2580
                    for(i=16; i<16+8; i++){
2581
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2582
                            uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
2583
                            svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2584
                        }
2585
                    }
2586
                }
2587
            }
2588
        }
2589
    }
2590
    if(h->cbp || IS_INTRA(mb_type))
2591
        s->dsp.clear_blocks(h->mb);
2592

    
2593
    if(h->deblocking_filter) {
2594
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
2595
        fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
2596
        h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2597
        h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2598
        if (!simple && FRAME_MBAFF) {
2599
            filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2600
        } else {
2601
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2602
        }
2603
    }
2604
}
2605

    
2606
/**
2607
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2608
 */
2609
static void hl_decode_mb_simple(H264Context *h){
2610
    hl_decode_mb_internal(h, 1);
2611
}
2612

    
2613
/**
2614
 * Process a macroblock; this handles edge cases, such as interlacing.
2615
 */
2616
static void av_noinline hl_decode_mb_complex(H264Context *h){
2617
    hl_decode_mb_internal(h, 0);
2618
}
2619

    
2620
static void hl_decode_mb(H264Context *h){
2621
    MpegEncContext * const s = &h->s;
2622
    const int mb_xy= h->mb_xy;
2623
    const int mb_type= s->current_picture.mb_type[mb_xy];
2624
    int is_complex = ENABLE_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2625

    
2626
    if(ENABLE_H264_ENCODER && !s->decode)
2627
        return;
2628

    
2629
    if (is_complex)
2630
        hl_decode_mb_complex(h);
2631
    else hl_decode_mb_simple(h);
2632
}
2633

    
2634
static void pic_as_field(Picture *pic, const int parity){
2635
    int i;
2636
    for (i = 0; i < 4; ++i) {
2637
        if (parity == PICT_BOTTOM_FIELD)
2638
            pic->data[i] += pic->linesize[i];
2639
        pic->reference = parity;
2640
        pic->linesize[i] *= 2;
2641
    }
2642
    pic->poc= pic->field_poc[parity == PICT_BOTTOM_FIELD];
2643
}
2644

    
2645
static int split_field_copy(Picture *dest, Picture *src,
2646
                            int parity, int id_add){
2647
    int match = !!(src->reference & parity);
2648

    
2649
    if (match) {
2650
        *dest = *src;
2651
        if(parity != PICT_FRAME){
2652
            pic_as_field(dest, parity);
2653
            dest->pic_id *= 2;
2654
            dest->pic_id += id_add;
2655
        }
2656
    }
2657

    
2658
    return match;
2659
}
2660

    
2661
static int build_def_list(Picture *def, Picture **in, int len, int is_long, int sel){
2662
    int i[2]={0};
2663
    int index=0;
2664

    
2665
    while(i[0]<len || i[1]<len){
2666
        while(i[0]<len && !(in[ i[0] ] && (in[ i[0] ]->reference & sel)))
2667
            i[0]++;
2668
        while(i[1]<len && !(in[ i[1] ] && (in[ i[1] ]->reference & (sel^3))))
2669
            i[1]++;
2670
        if(i[0] < len){
2671
            in[ i[0] ]->pic_id= is_long ? i[0] : in[ i[0] ]->frame_num;
2672
            split_field_copy(&def[index++], in[ i[0]++ ], sel  , 1);
2673
        }
2674
        if(i[1] < len){
2675
            in[ i[1] ]->pic_id= is_long ? i[1] : in[ i[1] ]->frame_num;
2676
            split_field_copy(&def[index++], in[ i[1]++ ], sel^3, 0);
2677
        }
2678
    }
2679

    
2680
    return index;
2681
}
2682

    
2683
static int add_sorted(Picture **sorted, Picture **src, int len, int limit, int dir){
2684
    int i, best_poc;
2685
    int out_i= 0;
2686

    
2687
    for(;;){
2688
        best_poc= dir ? INT_MIN : INT_MAX;
2689

    
2690
        for(i=0; i<len; i++){
2691
            const int poc= src[i]->poc;
2692
            if(((poc > limit) ^ dir) && ((poc < best_poc) ^ dir)){
2693
                best_poc= poc;
2694
                sorted[out_i]= src[i];
2695
            }
2696
        }
2697
        if(best_poc == (dir ? INT_MIN : INT_MAX))
2698
            break;
2699
        limit= sorted[out_i++]->poc - dir;
2700
    }
2701
    return out_i;
2702
}
2703

    
2704
/**
2705
 * fills the default_ref_list.
2706
 */
2707
static int fill_default_ref_list(H264Context *h){
2708
    MpegEncContext * const s = &h->s;
2709
    int i, len;
2710

    
2711
    if(h->slice_type_nos==FF_B_TYPE){
2712
        Picture *sorted[32];
2713
        int cur_poc, list;
2714
        int lens[2];
2715

    
2716
        if(FIELD_PICTURE)
2717
            cur_poc= s->current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
2718
        else
2719
            cur_poc= s->current_picture_ptr->poc;
2720

    
2721
        for(list= 0; list<2; list++){
2722
            len= add_sorted(sorted    , h->short_ref, h->short_ref_count, cur_poc, 1^list);
2723
            len+=add_sorted(sorted+len, h->short_ref, h->short_ref_count, cur_poc, 0^list);
2724
            assert(len<=32);
2725
            len= build_def_list(h->default_ref_list[list]    , sorted     , len, 0, s->picture_structure);
2726
            len+=build_def_list(h->default_ref_list[list]+len, h->long_ref, 16 , 1, s->picture_structure);
2727
            assert(len<=32);
2728

    
2729
            if(len < h->ref_count[list])
2730
                memset(&h->default_ref_list[list][len], 0, sizeof(Picture)*(h->ref_count[list] - len));
2731
            lens[list]= len;
2732
        }
2733

    
2734
        if(lens[0] == lens[1] && lens[1] > 1){
2735
            for(i=0; h->default_ref_list[0][i].data[0] == h->default_ref_list[1][i].data[0] && i<lens[0]; i++);
2736
            if(i == lens[0])
2737
                FFSWAP(Picture, h->default_ref_list[1][0], h->default_ref_list[1][1]);
2738
        }
2739
    }else{
2740
        len = build_def_list(h->default_ref_list[0]    , h->short_ref, h->short_ref_count, 0, s->picture_structure);
2741
        len+= build_def_list(h->default_ref_list[0]+len, h-> long_ref, 16                , 1, s->picture_structure);
2742
        assert(len <= 32);
2743
        if(len < h->ref_count[0])
2744
            memset(&h->default_ref_list[0][len], 0, sizeof(Picture)*(h->ref_count[0] - len));
2745
    }
2746
#ifdef TRACE
2747
    for (i=0; i<h->ref_count[0]; i++) {
2748
        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]);
2749
    }
2750
    if(h->slice_type_nos==FF_B_TYPE){
2751
        for (i=0; i<h->ref_count[1]; i++) {
2752
            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[1][i].data[0]);
2753
        }
2754
    }
2755
#endif
2756
    return 0;
2757
}
2758

    
2759
static void print_short_term(H264Context *h);
2760
static void print_long_term(H264Context *h);
2761

    
2762
/**
2763
 * Extract structure information about the picture described by pic_num in
2764
 * the current decoding context (frame or field). Note that pic_num is
2765
 * picture number without wrapping (so, 0<=pic_num<max_pic_num).
2766
 * @param pic_num picture number for which to extract structure information
2767
 * @param structure one of PICT_XXX describing structure of picture
2768
 *                      with pic_num
2769
 * @return frame number (short term) or long term index of picture
2770
 *         described by pic_num
2771
 */
2772
static int pic_num_extract(H264Context *h, int pic_num, int *structure){
2773
    MpegEncContext * const s = &h->s;
2774

    
2775
    *structure = s->picture_structure;
2776
    if(FIELD_PICTURE){
2777
        if (!(pic_num & 1))
2778
            /* opposite field */
2779
            *structure ^= PICT_FRAME;
2780
        pic_num >>= 1;
2781
    }
2782

    
2783
    return pic_num;
2784
}
2785

    
2786
static int decode_ref_pic_list_reordering(H264Context *h){
2787
    MpegEncContext * const s = &h->s;
2788
    int list, index, pic_structure;
2789

    
2790
    print_short_term(h);
2791
    print_long_term(h);
2792

    
2793
    for(list=0; list<h->list_count; list++){
2794
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2795

    
2796
        if(get_bits1(&s->gb)){
2797
            int pred= h->curr_pic_num;
2798

    
2799
            for(index=0; ; index++){
2800
                unsigned int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2801
                unsigned int pic_id;
2802
                int i;
2803
                Picture *ref = NULL;
2804

    
2805
                if(reordering_of_pic_nums_idc==3)
2806
                    break;
2807

    
2808
                if(index >= h->ref_count[list]){
2809
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2810
                    return -1;
2811
                }
2812

    
2813
                if(reordering_of_pic_nums_idc<3){
2814
                    if(reordering_of_pic_nums_idc<2){
2815
                        const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2816
                        int frame_num;
2817

    
2818
                        if(abs_diff_pic_num > h->max_pic_num){
2819
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2820
                            return -1;
2821
                        }
2822

    
2823
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2824
                        else                                pred+= abs_diff_pic_num;
2825
                        pred &= h->max_pic_num - 1;
2826

    
2827
                        frame_num = pic_num_extract(h, pred, &pic_structure);
2828

    
2829
                        for(i= h->short_ref_count-1; i>=0; i--){
2830
                            ref = h->short_ref[i];
2831
                            assert(ref->reference);
2832
                            assert(!ref->long_ref);
2833
                            if(
2834
                                   ref->frame_num == frame_num &&
2835
                                   (ref->reference & pic_structure)
2836
                              )
2837
                                break;
2838
                        }
2839
                        if(i>=0)
2840
                            ref->pic_id= pred;
2841
                    }else{
2842
                        int long_idx;
2843
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2844

    
2845
                        long_idx= pic_num_extract(h, pic_id, &pic_structure);
2846

    
2847
                        if(long_idx>31){
2848
                            av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
2849
                            return -1;
2850
                        }
2851
                        ref = h->long_ref[long_idx];
2852
                        assert(!(ref && !ref->reference));
2853
                        if(ref && (ref->reference & pic_structure)){
2854
                            ref->pic_id= pic_id;
2855
                            assert(ref->long_ref);
2856
                            i=0;
2857
                        }else{
2858
                            i=-1;
2859
                        }
2860
                    }
2861

    
2862
                    if (i < 0) {
2863
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2864
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2865
                    } else {
2866
                        for(i=index; i+1<h->ref_count[list]; i++){
2867
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
2868
                                break;
2869
                        }
2870
                        for(; i > index; i--){
2871
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2872
                        }
2873
                        h->ref_list[list][index]= *ref;
2874
                        if (FIELD_PICTURE){
2875
                            pic_as_field(&h->ref_list[list][index], pic_structure);
2876
                        }
2877
                    }
2878
                }else{
2879
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2880
                    return -1;
2881
                }
2882
            }
2883
        }
2884
    }
2885
    for(list=0; list<h->list_count; list++){
2886
        for(index= 0; index < h->ref_count[list]; index++){
2887
            if(!h->ref_list[list][index].data[0]){
2888
                av_log(h->s.avctx, AV_LOG_ERROR, "Missing reference picture\n");
2889
                h->ref_list[list][index]= s->current_picture; //FIXME this is not a sensible solution
2890
            }
2891
        }
2892
    }
2893

    
2894
    return 0;
2895
}
2896

    
2897
static void fill_mbaff_ref_list(H264Context *h){
2898
    int list, i, j;
2899
    for(list=0; list<2; list++){ //FIXME try list_count
2900
        for(i=0; i<h->ref_count[list]; i++){
2901
            Picture *frame = &h->ref_list[list][i];
2902
            Picture *field = &h->ref_list[list][16+2*i];
2903
            field[0] = *frame;
2904
            for(j=0; j<3; j++)
2905
                field[0].linesize[j] <<= 1;
2906
            field[0].reference = PICT_TOP_FIELD;
2907
            field[0].poc= field[0].field_poc[0];
2908
            field[1] = field[0];
2909
            for(j=0; j<3; j++)
2910
                field[1].data[j] += frame->linesize[j];
2911
            field[1].reference = PICT_BOTTOM_FIELD;
2912
            field[1].poc= field[1].field_poc[1];
2913

    
2914
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
2915
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
2916
            for(j=0; j<2; j++){
2917
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
2918
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
2919
            }
2920
        }
2921
    }
2922
    for(j=0; j<h->ref_count[1]; j++){
2923
        for(i=0; i<h->ref_count[0]; i++)
2924
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
2925
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
2926
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
2927
    }
2928
}
2929

    
2930
static int pred_weight_table(H264Context *h){
2931
    MpegEncContext * const s = &h->s;
2932
    int list, i;
2933
    int luma_def, chroma_def;
2934

    
2935
    h->use_weight= 0;
2936
    h->use_weight_chroma= 0;
2937
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2938
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2939
    luma_def = 1<<h->luma_log2_weight_denom;
2940
    chroma_def = 1<<h->chroma_log2_weight_denom;
2941

    
2942
    for(list=0; list<2; list++){
2943
        for(i=0; i<h->ref_count[list]; i++){
2944
            int luma_weight_flag, chroma_weight_flag;
2945

    
2946
            luma_weight_flag= get_bits1(&s->gb);
2947
            if(luma_weight_flag){
2948
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
2949
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
2950
                if(   h->luma_weight[list][i] != luma_def
2951
                   || h->luma_offset[list][i] != 0)
2952
                    h->use_weight= 1;
2953
            }else{
2954
                h->luma_weight[list][i]= luma_def;
2955
                h->luma_offset[list][i]= 0;
2956
            }
2957

    
2958
            if(CHROMA){
2959
                chroma_weight_flag= get_bits1(&s->gb);
2960
                if(chroma_weight_flag){
2961
                    int j;
2962
                    for(j=0; j<2; j++){
2963
                        h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2964
                        h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2965
                        if(   h->chroma_weight[list][i][j] != chroma_def
2966
                        || h->chroma_offset[list][i][j] != 0)
2967
                            h->use_weight_chroma= 1;
2968
                    }
2969
                }else{
2970
                    int j;
2971
                    for(j=0; j<2; j++){
2972
                        h->chroma_weight[list][i][j]= chroma_def;
2973
                        h->chroma_offset[list][i][j]= 0;
2974
                    }
2975
                }
2976
            }
2977
        }
2978
        if(h->slice_type_nos != FF_B_TYPE) break;
2979
    }
2980
    h->use_weight= h->use_weight || h->use_weight_chroma;
2981
    return 0;
2982
}
2983

    
2984
static void implicit_weight_table(H264Context *h){
2985
    MpegEncContext * const s = &h->s;
2986
    int ref0, ref1;
2987
    int cur_poc = s->current_picture_ptr->poc;
2988

    
2989
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
2990
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2991
        h->use_weight= 0;
2992
        h->use_weight_chroma= 0;
2993
        return;
2994
    }
2995

    
2996
    h->use_weight= 2;
2997
    h->use_weight_chroma= 2;
2998
    h->luma_log2_weight_denom= 5;
2999
    h->chroma_log2_weight_denom= 5;
3000

    
3001
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3002
        int poc0 = h->ref_list[0][ref0].poc;
3003
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3004
            int poc1 = h->ref_list[1][ref1].poc;
3005
            int td = av_clip(poc1 - poc0, -128, 127);
3006
            if(td){
3007
                int tb = av_clip(cur_poc - poc0, -128, 127);
3008
                int tx = (16384 + (FFABS(td) >> 1)) / td;
3009
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3010
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3011
                    h->implicit_weight[ref0][ref1] = 32;
3012
                else
3013
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3014
            }else
3015
                h->implicit_weight[ref0][ref1] = 32;
3016
        }
3017
    }
3018
}
3019

    
3020
/**
3021
 * Mark a picture as no longer needed for reference. The refmask
3022
 * argument allows unreferencing of individual fields or the whole frame.
3023
 * If the picture becomes entirely unreferenced, but is being held for
3024
 * display purposes, it is marked as such.
3025
 * @param refmask mask of fields to unreference; the mask is bitwise
3026
 *                anded with the reference marking of pic
3027
 * @return non-zero if pic becomes entirely unreferenced (except possibly
3028
 *         for display purposes) zero if one of the fields remains in
3029
 *         reference
3030
 */
3031
static inline int unreference_pic(H264Context *h, Picture *pic, int refmask){
3032
    int i;
3033
    if (pic->reference &= refmask) {
3034
        return 0;
3035
    } else {
3036
        for(i = 0; h->delayed_pic[i]; i++)
3037
            if(pic == h->delayed_pic[i]){
3038
                pic->reference=DELAYED_PIC_REF;
3039
                break;
3040
            }
3041
        return 1;
3042
    }
3043
}
3044

    
3045
/**
3046
 * instantaneous decoder refresh.
3047
 */
3048
static void idr(H264Context *h){
3049
    int i;
3050

    
3051
    for(i=0; i<16; i++){
3052
        remove_long(h, i, 0);
3053
    }
3054
    assert(h->long_ref_count==0);
3055

    
3056
    for(i=0; i<h->short_ref_count; i++){
3057
        unreference_pic(h, h->short_ref[i], 0);
3058
        h->short_ref[i]= NULL;
3059
    }
3060
    h->short_ref_count=0;
3061
    h->prev_frame_num= 0;
3062
    h->prev_frame_num_offset= 0;
3063
    h->prev_poc_msb=
3064
    h->prev_poc_lsb= 0;
3065
}
3066

    
3067
/* forget old pics after a seek */
3068
static void flush_dpb(AVCodecContext *avctx){
3069
    H264Context *h= avctx->priv_data;
3070
    int i;
3071
    for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
3072
        if(h->delayed_pic[i])
3073
            h->delayed_pic[i]->reference= 0;
3074
        h->delayed_pic[i]= NULL;
3075
    }
3076
    h->outputed_poc= INT_MIN;
3077
    idr(h);
3078
    if(h->s.current_picture_ptr)
3079
        h->s.current_picture_ptr->reference= 0;
3080
    h->s.first_field= 0;
3081
    ff_mpeg_flush(avctx);
3082
}
3083

    
3084
/**
3085
 * Find a Picture in the short term reference list by frame number.
3086
 * @param frame_num frame number to search for
3087
 * @param idx the index into h->short_ref where returned picture is found
3088
 *            undefined if no picture found.
3089
 * @return pointer to the found picture, or NULL if no pic with the provided
3090
 *                 frame number is found
3091
 */
3092
static Picture * find_short(H264Context *h, int frame_num, int *idx){
3093
    MpegEncContext * const s = &h->s;
3094
    int i;
3095

    
3096
    for(i=0; i<h->short_ref_count; i++){
3097
        Picture *pic= h->short_ref[i];
3098
        if(s->avctx->debug&FF_DEBUG_MMCO)
3099
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3100
        if(pic->frame_num == frame_num) {
3101
            *idx = i;
3102
            return pic;
3103
        }
3104
    }
3105
    return NULL;
3106
}
3107

    
3108
/**
3109
 * Remove a picture from the short term reference list by its index in
3110
 * that list.  This does no checking on the provided index; it is assumed
3111
 * to be valid. Other list entries are shifted down.
3112
 * @param i index into h->short_ref of picture to remove.
3113
 */
3114
static void remove_short_at_index(H264Context *h, int i){
3115
    assert(i >= 0 && i < h->short_ref_count);
3116
    h->short_ref[i]= NULL;
3117
    if (--h->short_ref_count)
3118
        memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i)*sizeof(Picture*));
3119
}
3120

    
3121
/**
3122
 *
3123
 * @return the removed picture or NULL if an error occurs
3124
 */
3125
static Picture * remove_short(H264Context *h, int frame_num, int ref_mask){
3126
    MpegEncContext * const s = &h->s;
3127
    Picture *pic;
3128
    int i;
3129

    
3130
    if(s->avctx->debug&FF_DEBUG_MMCO)
3131
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3132

    
3133
    pic = find_short(h, frame_num, &i);
3134
    if (pic){
3135
        if(unreference_pic(h, pic, ref_mask))
3136
        remove_short_at_index(h, i);
3137
    }
3138

    
3139
    return pic;
3140
}
3141

    
3142
/**
3143
 * Remove a picture from the long term reference list by its index in
3144
 * that list.
3145
 * @return the removed picture or NULL if an error occurs
3146
 */
3147
static Picture * remove_long(H264Context *h, int i, int ref_mask){
3148
    Picture *pic;
3149

    
3150
    pic= h->long_ref[i];
3151
    if (pic){
3152
        if(unreference_pic(h, pic, ref_mask)){
3153
            assert(h->long_ref[i]->long_ref == 1);
3154
            h->long_ref[i]->long_ref= 0;
3155
            h->long_ref[i]= NULL;
3156
            h->long_ref_count--;
3157
        }
3158
    }
3159

    
3160
    return pic;
3161
}
3162

    
3163
/**
3164
 * print short term list
3165
 */
3166
static void print_short_term(H264Context *h) {
3167
    uint32_t i;
3168
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3169
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3170
        for(i=0; i<h->short_ref_count; i++){
3171
            Picture *pic= h->short_ref[i];
3172
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3173
        }
3174
    }
3175
}
3176

    
3177
/**
3178
 * print long term list
3179
 */
3180
static void print_long_term(H264Context *h) {
3181
    uint32_t i;
3182
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3183
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3184
        for(i = 0; i < 16; i++){
3185
            Picture *pic= h->long_ref[i];
3186
            if (pic) {
3187
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3188
            }
3189
        }
3190
    }
3191
}
3192

    
3193
/**
3194
 * Executes the reference picture marking (memory management control operations).
3195
 */
3196
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3197
    MpegEncContext * const s = &h->s;
3198
    int i, j;
3199
    int current_ref_assigned=0;
3200
    Picture *pic;
3201

    
3202
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3203
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3204

    
3205
    for(i=0; i<mmco_count; i++){
3206
        int structure, frame_num;
3207
        if(s->avctx->debug&FF_DEBUG_MMCO)
3208
            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);
3209

    
3210
        if(   mmco[i].opcode == MMCO_SHORT2UNUSED
3211
           || mmco[i].opcode == MMCO_SHORT2LONG){
3212
            frame_num = pic_num_extract(h, mmco[i].short_pic_num, &structure);
3213
            pic = find_short(h, frame_num, &j);
3214
            if(!pic){
3215
                if(mmco[i].opcode != MMCO_SHORT2LONG || !h->long_ref[mmco[i].long_arg]
3216
                   || h->long_ref[mmco[i].long_arg]->frame_num != frame_num)
3217
                av_log(h->s.avctx, AV_LOG_ERROR, "mmco: unref short failure\n");
3218
                continue;
3219
            }
3220
        }
3221

    
3222
        switch(mmco[i].opcode){
3223
        case MMCO_SHORT2UNUSED:
3224
            if(s->avctx->debug&FF_DEBUG_MMCO)
3225
                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);
3226
            remove_short(h, frame_num, structure ^ PICT_FRAME);
3227
            break;
3228
        case MMCO_SHORT2LONG:
3229
                if (h->long_ref[mmco[i].long_arg] != pic)
3230
                    remove_long(h, mmco[i].long_arg, 0);
3231

    
3232
                remove_short_at_index(h, j);
3233
                h->long_ref[ mmco[i].long_arg ]= pic;
3234
                if (h->long_ref[ mmco[i].long_arg ]){
3235
                    h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3236
                    h->long_ref_count++;
3237
                }
3238
            break;
3239
        case MMCO_LONG2UNUSED:
3240
            j = pic_num_extract(h, mmco[i].long_arg, &structure);
3241
            pic = h->long_ref[j];
3242
            if (pic) {
3243
                remove_long(h, j, structure ^ PICT_FRAME);
3244
            } else if(s->avctx->debug&FF_DEBUG_MMCO)
3245
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref long failure\n");
3246
            break;
3247
        case MMCO_LONG:
3248
                    // Comment below left from previous code as it is an interresting note.
3249
                    /* First field in pair is in short term list or
3250
                     * at a different long term index.
3251
                     * This is not allowed; see 7.4.3.3, notes 2 and 3.
3252
                     * Report the problem and keep the pair where it is,
3253
                     * and mark this field valid.
3254
                     */
3255

    
3256
            if (h->long_ref[mmco[i].long_arg] != s->current_picture_ptr) {
3257
                remove_long(h, mmco[i].long_arg, 0);
3258

    
3259
                h->long_ref[ mmco[i].long_arg ]= s->current_picture_ptr;
3260
                h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3261
                h->long_ref_count++;
3262
            }
3263

    
3264
            s->current_picture_ptr->reference |= s->picture_structure;
3265
            current_ref_assigned=1;
3266
            break;
3267
        case MMCO_SET_MAX_LONG:
3268
            assert(mmco[i].long_arg <= 16);
3269
            // just remove the long term which index is greater than new max
3270
            for(j = mmco[i].long_arg; j<16; j++){
3271
                remove_long(h, j, 0);
3272
            }
3273
            break;
3274
        case MMCO_RESET:
3275
            while(h->short_ref_count){
3276
                remove_short(h, h->short_ref[0]->frame_num, 0);
3277
            }
3278
            for(j = 0; j < 16; j++) {
3279
                remove_long(h, j, 0);
3280
            }
3281
            s->current_picture_ptr->poc=
3282
            s->current_picture_ptr->field_poc[0]=
3283
            s->current_picture_ptr->field_poc[1]=
3284
            h->poc_lsb=
3285
            h->poc_msb=
3286
            h->frame_num=
3287
            s->current_picture_ptr->frame_num= 0;
3288
            break;
3289
        default: assert(0);
3290
        }
3291
    }
3292

    
3293
    if (!current_ref_assigned) {
3294
        /* Second field of complementary field pair; the first field of
3295
         * which is already referenced. If short referenced, it
3296
         * should be first entry in short_ref. If not, it must exist
3297
         * in long_ref; trying to put it on the short list here is an
3298
         * error in the encoded bit stream (ref: 7.4.3.3, NOTE 2 and 3).
3299
         */
3300
        if (h->short_ref_count && h->short_ref[0] == s->current_picture_ptr) {
3301
            /* Just mark the second field valid */
3302
            s->current_picture_ptr->reference = PICT_FRAME;
3303
        } else if (s->current_picture_ptr->long_ref) {
3304
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term reference "
3305
                                             "assignment for second field "
3306
                                             "in complementary field pair "
3307
                                             "(first field is long term)\n");
3308
        } else {
3309
            pic= remove_short(h, s->current_picture_ptr->frame_num, 0);
3310
            if(pic){
3311
                av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3312
            }
3313

    
3314
            if(h->short_ref_count)
3315
                memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3316

    
3317
            h->short_ref[0]= s->current_picture_ptr;
3318
            h->short_ref_count++;
3319
            s->current_picture_ptr->reference |= s->picture_structure;
3320
        }
3321
    }
3322

    
3323
    if (h->long_ref_count + h->short_ref_count > h->sps.ref_frame_count){
3324

    
3325
        /* We have too many reference frames, probably due to corrupted
3326
         * stream. Need to discard one frame. Prevents overrun of the
3327
         * short_ref and long_ref buffers.
3328
         */
3329
        av_log(h->s.avctx, AV_LOG_ERROR,
3330
               "number of reference frames exceeds max (probably "
3331
               "corrupt input), discarding one\n");
3332

    
3333
        if (h->long_ref_count && !h->short_ref_count) {
3334
            for (i = 0; i < 16; ++i)
3335
                if (h->long_ref[i])
3336
                    break;
3337

    
3338
            assert(i < 16);
3339
            remove_long(h, i, 0);
3340
        } else {
3341
            pic = h->short_ref[h->short_ref_count - 1];
3342
            remove_short(h, pic->frame_num, 0);
3343
        }
3344
    }
3345

    
3346
    print_short_term(h);
3347
    print_long_term(h);
3348
    return 0;
3349
}
3350

    
3351
static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
3352
    MpegEncContext * const s = &h->s;
3353
    int i;
3354

    
3355
    h->mmco_index= 0;
3356
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3357
        s->broken_link= get_bits1(gb) -1;
3358
        if(get_bits1(gb)){
3359
            h->mmco[0].opcode= MMCO_LONG;
3360
            h->mmco[0].long_arg= 0;
3361
            h->mmco_index= 1;
3362
        }
3363
    }else{
3364
        if(get_bits1(gb)){ // adaptive_ref_pic_marking_mode_flag
3365
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
3366
                MMCOOpcode opcode= get_ue_golomb(gb);
3367

    
3368
                h->mmco[i].opcode= opcode;
3369
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3370
                    h->mmco[i].short_pic_num= (h->curr_pic_num - get_ue_golomb(gb) - 1) & (h->max_pic_num - 1);
3371
/*                    if(h->mmco[i].short_pic_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_pic_num ] == NULL){
3372
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
3373
                        return -1;
3374
                    }*/
3375
                }
3376
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3377
                    unsigned int long_arg= get_ue_golomb(gb);
3378
                    if(long_arg >= 32 || (long_arg >= 16 && !(opcode == MMCO_LONG2UNUSED && FIELD_PICTURE))){
3379
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3380
                        return -1;
3381
                    }
3382
                    h->mmco[i].long_arg= long_arg;
3383
                }
3384

    
3385
                if(opcode > (unsigned)MMCO_LONG){
3386
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3387
                    return -1;
3388
                }
3389
                if(opcode == MMCO_END)
3390
                    break;
3391
            }
3392
            h->mmco_index= i;
3393
        }else{
3394
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3395

    
3396
            if(h->short_ref_count && h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count &&
3397
                    !(FIELD_PICTURE && !s->first_field && s->current_picture_ptr->reference)) {
3398
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3399
                h->mmco[0].short_pic_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3400
                h->mmco_index= 1;
3401
                if (FIELD_PICTURE) {
3402
                    h->mmco[0].short_pic_num *= 2;
3403
                    h->mmco[1].opcode= MMCO_SHORT2UNUSED;
3404
                    h->mmco[1].short_pic_num= h->mmco[0].short_pic_num + 1;
3405
                    h->mmco_index= 2;
3406
                }
3407
            }
3408
        }
3409
    }
3410

    
3411
    return 0;
3412
}
3413

    
3414
static int init_poc(H264Context *h){
3415
    MpegEncContext * const s = &h->s;
3416
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3417
    int field_poc[2];
3418
    Picture *cur = s->current_picture_ptr;
3419

    
3420
    h->frame_num_offset= h->prev_frame_num_offset;
3421
    if(h->frame_num < h->prev_frame_num)
3422
        h->frame_num_offset += max_frame_num;
3423

    
3424
    if(h->sps.poc_type==0){
3425
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3426

    
3427
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3428
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3429
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3430
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3431
        else
3432
            h->poc_msb = h->prev_poc_msb;
3433
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3434
        field_poc[0] =
3435
        field_poc[1] = h->poc_msb + h->poc_lsb;
3436
        if(s->picture_structure == PICT_FRAME)
3437
            field_poc[1] += h->delta_poc_bottom;
3438
    }else if(h->sps.poc_type==1){
3439
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3440
        int i;
3441

    
3442
        if(h->sps.poc_cycle_length != 0)
3443
            abs_frame_num = h->frame_num_offset + h->frame_num;
3444
        else
3445
            abs_frame_num = 0;
3446

    
3447
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
3448
            abs_frame_num--;
3449

    
3450
        expected_delta_per_poc_cycle = 0;
3451
        for(i=0; i < h->sps.poc_cycle_length; i++)
3452
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3453

    
3454
        if(abs_frame_num > 0){
3455
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3456
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3457

    
3458
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3459
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
3460
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3461
        } else
3462
            expectedpoc = 0;
3463

    
3464
        if(h->nal_ref_idc == 0)
3465
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3466

    
3467
        field_poc[0] = expectedpoc + h->delta_poc[0];
3468
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3469

    
3470
        if(s->picture_structure == PICT_FRAME)
3471
            field_poc[1] += h->delta_poc[1];
3472
    }else{
3473
        int poc= 2*(h->frame_num_offset + h->frame_num);
3474

    
3475
        if(!h->nal_ref_idc)
3476
            poc--;
3477

    
3478
        field_poc[0]= poc;
3479
        field_poc[1]= poc;
3480
    }
3481

    
3482
    if(s->picture_structure != PICT_BOTTOM_FIELD)
3483
        s->current_picture_ptr->field_poc[0]= field_poc[0];
3484
    if(s->picture_structure != PICT_TOP_FIELD)
3485
        s->current_picture_ptr->field_poc[1]= field_poc[1];
3486
    cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
3487

    
3488
    return 0;
3489
}
3490

    
3491

    
3492
/**
3493
 * initialize scan tables
3494
 */
3495
static void init_scan_tables(H264Context *h){
3496
    MpegEncContext * const s = &h->s;
3497
    int i;
3498
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
3499
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
3500
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
3501
    }else{
3502
        for(i=0; i<16; i++){
3503
#define T(x) (x>>2) | ((x<<2) & 0xF)
3504
            h->zigzag_scan[i] = T(zigzag_scan[i]);
3505
            h-> field_scan[i] = T( field_scan[i]);
3506
#undef T
3507
        }
3508
    }
3509
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
3510
        memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));
3511
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
3512
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
3513
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
3514
    }else{
3515
        for(i=0; i<64; i++){
3516
#define T(x) (x>>3) | ((x&7)<<3)
3517
            h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);
3518
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
3519
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
3520
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
3521
#undef T
3522
        }
3523
    }
3524
    if(h->sps.transform_bypass){ //FIXME same ugly
3525
        h->zigzag_scan_q0          = zigzag_scan;
3526
        h->zigzag_scan8x8_q0       = zigzag_scan8x8;
3527
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
3528
        h->field_scan_q0           = field_scan;
3529
        h->field_scan8x8_q0        = field_scan8x8;
3530
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
3531
    }else{
3532
        h->zigzag_scan_q0          = h->zigzag_scan;
3533
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
3534
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
3535
        h->field_scan_q0           = h->field_scan;
3536
        h->field_scan8x8_q0        = h->field_scan8x8;
3537
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
3538
    }
3539
}
3540

    
3541
/**
3542
 * Replicates H264 "master" context to thread contexts.
3543
 */
3544
static void clone_slice(H264Context *dst, H264Context *src)
3545
{
3546
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
3547
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
3548
    dst->s.current_picture      = src->s.current_picture;
3549
    dst->s.linesize             = src->s.linesize;
3550
    dst->s.uvlinesize           = src->s.uvlinesize;
3551
    dst->s.first_field          = src->s.first_field;
3552

    
3553
    dst->prev_poc_msb           = src->prev_poc_msb;
3554
    dst->prev_poc_lsb           = src->prev_poc_lsb;
3555
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
3556
    dst->prev_frame_num         = src->prev_frame_num;
3557
    dst->short_ref_count        = src->short_ref_count;
3558

    
3559
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
3560
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
3561
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
3562
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
3563

    
3564
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
3565
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
3566
}
3567

    
3568
/**
3569
 * decodes a slice header.
3570
 * This will also call MPV_common_init() and frame_start() as needed.
3571
 *
3572
 * @param h h264context
3573
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
3574
 *
3575
 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3576
 */
3577
static int decode_slice_header(H264Context *h, H264Context *h0){
3578
    MpegEncContext * const s = &h->s;
3579
    MpegEncContext * const s0 = &h0->s;
3580
    unsigned int first_mb_in_slice;
3581
    unsigned int pps_id;
3582
    int num_ref_idx_active_override_flag;
3583
    unsigned int slice_type, tmp, i, j;
3584
    int default_ref_list_done = 0;
3585
    int last_pic_structure;
3586

    
3587
    s->dropable= h->nal_ref_idc == 0;
3588

    
3589
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
3590
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
3591
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
3592
    }else{
3593
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
3594
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
3595
    }
3596

    
3597
    first_mb_in_slice= get_ue_golomb(&s->gb);
3598

    
3599
    if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
3600
        h0->current_slice = 0;
3601
        if (!s0->first_field)
3602
            s->current_picture_ptr= NULL;
3603
    }
3604

    
3605
    slice_type= get_ue_golomb(&s->gb);
3606
    if(slice_type > 9){
3607
        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);
3608
        return -1;
3609
    }
3610
    if(slice_type > 4){
3611
        slice_type -= 5;
3612
        h->slice_type_fixed=1;
3613
    }else
3614
        h->slice_type_fixed=0;
3615

    
3616
    slice_type= golomb_to_pict_type[ slice_type ];
3617
    if (slice_type == FF_I_TYPE
3618
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
3619
        default_ref_list_done = 1;
3620
    }
3621
    h->slice_type= slice_type;
3622
    h->slice_type_nos= slice_type & 3;
3623

    
3624
    s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
3625
    if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
3626
        av_log(h->s.avctx, AV_LOG_ERROR,
3627
               "B picture before any references, skipping\n");
3628
        return -1;
3629
    }
3630

    
3631
    pps_id= get_ue_golomb(&s->gb);
3632
    if(pps_id>=MAX_PPS_COUNT){
3633
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3634
        return -1;
3635
    }
3636
    if(!h0->pps_buffers[pps_id]) {
3637
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS referenced\n");
3638
        return -1;
3639
    }
3640
    h->pps= *h0->pps_buffers[pps_id];
3641

    
3642
    if(!h0->sps_buffers[h->pps.sps_id]) {
3643
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS referenced\n");
3644
        return -1;
3645
    }
3646
    h->sps = *h0->sps_buffers[h->pps.sps_id];
3647

    
3648
    if(h == h0 && h->dequant_coeff_pps != pps_id){
3649
        h->dequant_coeff_pps = pps_id;
3650
        init_dequant_tables(h);
3651
    }
3652

    
3653
    s->mb_width= h->sps.mb_width;
3654
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3655

    
3656
    h->b_stride=  s->mb_width*4;
3657
    h->b8_stride= s->mb_width*2;
3658

    
3659
    s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
3660
    if(h->sps.frame_mbs_only_flag)
3661
        s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
3662
    else
3663
        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
3664

    
3665
    if (s->context_initialized
3666
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3667
        if(h != h0)
3668
            return -1;   // width / height changed during parallelized decoding
3669
        free_tables(h);
3670
        flush_dpb(s->avctx);
3671
        MPV_common_end(s);
3672
    }
3673
    if (!s->context_initialized) {
3674
        if(h != h0)
3675
            return -1;  // we cant (re-)initialize context during parallel decoding
3676
        if (MPV_common_init(s) < 0)
3677
            return -1;
3678
        s->first_field = 0;
3679

    
3680
        init_scan_tables(h);
3681
        alloc_tables(h);
3682

    
3683
        for(i = 1; i < s->avctx->thread_count; i++) {
3684
            H264Context *c;
3685
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
3686
            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
3687
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
3688
            c->sps = h->sps;
3689
            c->pps = h->pps;
3690
            init_scan_tables(c);
3691
            clone_tables(c, h);
3692
        }
3693

    
3694
        for(i = 0; i < s->avctx->thread_count; i++)
3695
            if(context_init(h->thread_context[i]) < 0)
3696
                return -1;
3697

    
3698
        s->avctx->width = s->width;
3699
        s->avctx->height = s->height;
3700
        s->avctx->sample_aspect_ratio= h->sps.sar;
3701
        if(!s->avctx->sample_aspect_ratio.den)
3702
            s->avctx->sample_aspect_ratio.den = 1;
3703

    
3704
        if(h->sps.timing_info_present_flag){
3705
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale};
3706
            if(h->x264_build > 0 && h->x264_build < 44)
3707
                s->avctx->time_base.den *= 2;
3708
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3709
                      s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
3710
        }
3711
    }
3712

    
3713
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3714

    
3715
    h->mb_mbaff = 0;
3716
    h->mb_aff_frame = 0;
3717
    last_pic_structure = s0->picture_structure;
3718
    if(h->sps.frame_mbs_only_flag){
3719
        s->picture_structure= PICT_FRAME;
3720
    }else{
3721
        if(get_bits1(&s->gb)) { //field_pic_flag
3722
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3723
        } else {
3724
            s->picture_structure= PICT_FRAME;
3725
            h->mb_aff_frame = h->sps.mb_aff;
3726
        }
3727
    }
3728
    h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
3729

    
3730
    if(h0->current_slice == 0){
3731
        while(h->frame_num !=  h->prev_frame_num &&
3732
              h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
3733
            av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
3734
            frame_start(h);
3735
            h->prev_frame_num++;
3736
            h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
3737
            s->current_picture_ptr->frame_num= h->prev_frame_num;
3738
            execute_ref_pic_marking(h, NULL, 0);
3739
        }
3740

    
3741
        /* See if we have a decoded first field looking for a pair... */
3742
        if (s0->first_field) {
3743
            assert(s0->current_picture_ptr);
3744
            assert(s0->current_picture_ptr->data[0]);
3745
            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
3746

    
3747
            /* figure out if we have a complementary field pair */
3748
            if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
3749
                /*
3750
                 * Previous field is unmatched. Don't display it, but let it
3751
                 * remain for reference if marked as such.
3752
                 */
3753
                s0->current_picture_ptr = NULL;
3754
                s0->first_field = FIELD_PICTURE;
3755

    
3756
            } else {
3757
                if (h->nal_ref_idc &&
3758
                        s0->current_picture_ptr->reference &&
3759
                        s0->current_picture_ptr->frame_num != h->frame_num) {
3760
                    /*
3761
                     * This and previous field were reference, but had
3762
                     * different frame_nums. Consider this field first in
3763
                     * pair. Throw away previous field except for reference
3764
                     * purposes.
3765
                     */
3766
                    s0->first_field = 1;
3767
                    s0->current_picture_ptr = NULL;
3768

    
3769
                } else {
3770
                    /* Second field in complementary pair */
3771
                    s0->first_field = 0;
3772
                }
3773
            }
3774

    
3775
        } else {
3776
            /* Frame or first field in a potentially complementary pair */
3777
            assert(!s0->current_picture_ptr);
3778
            s0->first_field = FIELD_PICTURE;
3779
        }
3780

    
3781
        if((!FIELD_PICTURE || s0->first_field) && frame_start(h) < 0) {
3782
            s0->first_field = 0;
3783
            return -1;
3784
        }
3785
    }
3786
    if(h != h0)
3787
        clone_slice(h, h0);
3788

    
3789
    s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
3790

    
3791
    assert(s->mb_num == s->mb_width * s->mb_height);
3792
    if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
3793
       first_mb_in_slice                    >= s->mb_num){
3794
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3795
        return -1;
3796
    }
3797
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3798
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
3799
    if (s->picture_structure == PICT_BOTTOM_FIELD)
3800
        s->resync_mb_y = s->mb_y = s->mb_y + 1;
3801
    assert(s->mb_y < s->mb_height);
3802

    
3803
    if(s->picture_structure==PICT_FRAME){
3804
        h->curr_pic_num=   h->frame_num;
3805
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3806
    }else{
3807
        h->curr_pic_num= 2*h->frame_num + 1;
3808
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3809
    }
3810

    
3811
    if(h->nal_unit_type == NAL_IDR_SLICE){
3812
        get_ue_golomb(&s->gb); /* idr_pic_id */
3813
    }
3814

    
3815
    if(h->sps.poc_type==0){
3816
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3817

    
3818
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3819
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3820
        }
3821
    }
3822

    
3823
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3824
        h->delta_poc[0]= get_se_golomb(&s->gb);
3825

    
3826
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3827
            h->delta_poc[1]= get_se_golomb(&s->gb);
3828
    }
3829

    
3830
    init_poc(h);
3831

    
3832
    if(h->pps.redundant_pic_cnt_present){
3833
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3834
    }
3835

    
3836
    //set defaults, might be overridden a few lines later
3837
    h->ref_count[0]= h->pps.ref_count[0];
3838
    h->ref_count[1]= h->pps.ref_count[1];
3839

    
3840
    if(h->slice_type_nos != FF_I_TYPE){
3841
        if(h->slice_type_nos == FF_B_TYPE){
3842
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3843
        }
3844
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3845

    
3846
        if(num_ref_idx_active_override_flag){
3847
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3848
            if(h->slice_type_nos==FF_B_TYPE)
3849
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3850

    
3851
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
3852
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3853
                h->ref_count[0]= h->ref_count[1]= 1;
3854
                return -1;
3855
            }
3856
        }
3857
        if(h->slice_type_nos == FF_B_TYPE)
3858
            h->list_count= 2;
3859
        else
3860
            h->list_count= 1;
3861
    }else
3862
        h->list_count= 0;
3863

    
3864
    if(!default_ref_list_done){
3865
        fill_default_ref_list(h);
3866
    }
3867

    
3868
    if(h->slice_type_nos!=FF_I_TYPE && decode_ref_pic_list_reordering(h) < 0)
3869
        return -1;
3870

    
3871
    if(h->slice_type_nos!=FF_I_TYPE){
3872
        s->last_picture_ptr= &h->ref_list[0][0];
3873
        ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3874
    }
3875
    if(h->slice_type_nos==FF_B_TYPE){
3876
        s->next_picture_ptr= &h->ref_list[1][0];
3877
        ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3878
    }
3879

    
3880
    if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
3881
       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
3882
        pred_weight_table(h);
3883
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
3884
        implicit_weight_table(h);
3885
    else
3886
        h->use_weight = 0;
3887

    
3888
    if(h->nal_ref_idc)
3889
        decode_ref_pic_marking(h0, &s->gb);
3890

    
3891
    if(FRAME_MBAFF)
3892
        fill_mbaff_ref_list(h);
3893

    
3894
    if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
3895
        direct_dist_scale_factor(h);
3896
    direct_ref_list_init(h);
3897

    
3898
    if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
3899
        tmp = get_ue_golomb(&s->gb);
3900
        if(tmp > 2){
3901
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3902
            return -1;
3903
        }
3904
        h->cabac_init_idc= tmp;
3905
    }
3906

    
3907
    h->last_qscale_diff = 0;
3908
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3909
    if(tmp>51){
3910
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3911
        return -1;
3912
    }
3913
    s->qscale= tmp;
3914
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3915
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3916
    //FIXME qscale / qp ... stuff
3917
    if(h->slice_type == FF_SP_TYPE){
3918
        get_bits1(&s->gb); /* sp_for_switch_flag */
3919
    }
3920
    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
3921
        get_se_golomb(&s->gb); /* slice_qs_delta */
3922
    }
3923

    
3924
    h->deblocking_filter = 1;
3925
    h->slice_alpha_c0_offset = 0;
3926
    h->slice_beta_offset = 0;
3927
    if( h->pps.deblocking_filter_parameters_present ) {
3928
        tmp= get_ue_golomb(&s->gb);
3929
        if(tmp > 2){
3930
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3931
            return -1;
3932
        }
3933
        h->deblocking_filter= tmp;
3934
        if(h->deblocking_filter < 2)
3935
            h->deblocking_filter^= 1; // 1<->0
3936

    
3937
        if( h->deblocking_filter ) {
3938
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3939
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3940
        }
3941
    }
3942

    
3943
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
3944
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
3945
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
3946
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3947
        h->deblocking_filter= 0;
3948

    
3949
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
3950
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3951
            /* Cheat slightly for speed:
3952
               Do not bother to deblock across slices. */
3953
            h->deblocking_filter = 2;
3954
        } else {
3955
            h0->max_contexts = 1;
3956
            if(!h0->single_decode_warning) {
3957
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3958
                h0->single_decode_warning = 1;
3959
            }
3960
            if(h != h0)
3961
                return 1; // deblocking switched inside frame
3962
        }
3963
    }
3964

    
3965
#if 0 //FMO
3966
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3967
        slice_group_change_cycle= get_bits(&s->gb, ?);
3968
#endif
3969

    
3970
    h0->last_slice_type = slice_type;
3971
    h->slice_num = ++h0->current_slice;
3972
    if(h->slice_num >= MAX_SLICES){
3973
        av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
3974
    }
3975

    
3976
    for(j=0; j<2; j++){
3977
        int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
3978
        ref2frm[0]=
3979
        ref2frm[1]= -1;
3980
        for(i=0; i<16; i++)
3981
            ref2frm[i+2]= 4*h->ref_list[j][i].frame_num
3982
                          +(h->ref_list[j][i].reference&3);
3983
        ref2frm[18+0]=
3984
        ref2frm[18+1]= -1;
3985
        for(i=16; i<48; i++)
3986
            ref2frm[i+4]= 4*h->ref_list[j][i].frame_num
3987
                          +(h->ref_list[j][i].reference&3);
3988
    }
3989

    
3990
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
3991
    h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3992

    
3993
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3994
        av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
3995
               h->slice_num,
3996
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3997
               first_mb_in_slice,
3998
               av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3999
               pps_id, h->frame_num,
4000
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4001
               h->ref_count[0], h->ref_count[1],
4002
               s->qscale,
4003
               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4004
               h->use_weight,
4005
               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
4006
               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
4007
               );
4008
    }
4009

    
4010
    return 0;
4011
}
4012

    
4013
/**
4014
 *
4015
 */
4016
static inline int get_level_prefix(GetBitContext *gb){
4017
    unsigned int buf;
4018
    int log;
4019

    
4020
    OPEN_READER(re, gb);
4021
    UPDATE_CACHE(re, gb);
4022
    buf=GET_CACHE(re, gb);
4023

    
4024
    log= 32 - av_log2(buf);
4025
#ifdef TRACE
4026
    print_bin(buf>>(32-log), log);
4027
    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__);
4028
#endif
4029

    
4030
    LAST_SKIP_BITS(re, gb, log);
4031
    CLOSE_READER(re, gb);
4032

    
4033
    return log-1;
4034
}
4035

    
4036
static inline int get_dct8x8_allowed(H264Context *h){
4037
    if(h->sps.direct_8x8_inference_flag)
4038
        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
4039
    else
4040
        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
4041
}
4042

    
4043
/**
4044
 * decodes a residual block.
4045
 * @param n block index
4046
 * @param scantable scantable
4047
 * @param max_coeff number of coefficients in the block
4048
 * @return <0 if an error occurred
4049
 */
4050
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
4051
    MpegEncContext * const s = &h->s;
4052
    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};
4053
    int level[16];
4054
    int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;
4055

    
4056
    //FIXME put trailing_onex into the context
4057

    
4058
    if(n == CHROMA_DC_BLOCK_INDEX){
4059
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4060
        total_coeff= coeff_token>>2;
4061
    }else{
4062
        if(n == LUMA_DC_BLOCK_INDEX){
4063
            total_coeff= pred_non_zero_count(h, 0);
4064
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4065
            total_coeff= coeff_token>>2;
4066
        }else{
4067
            total_coeff= pred_non_zero_count(h, n);
4068
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4069
            total_coeff= coeff_token>>2;
4070
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4071
        }
4072
    }
4073

    
4074
    //FIXME set last_non_zero?
4075

    
4076
    if(total_coeff==0)
4077
        return 0;
4078
    if(total_coeff > (unsigned)max_coeff) {
4079
        av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
4080
        return -1;
4081
    }
4082

    
4083
    trailing_ones= coeff_token&3;
4084
    tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
4085
    assert(total_coeff<=16);
4086

    
4087
    i = show_bits(gb, 3);
4088
    skip_bits(gb, trailing_ones);
4089
    level[0] = 1-((i&4)>>1);
4090
    level[1] = 1-((i&2)   );
4091
    level[2] = 1-((i&1)<<1);
4092

    
4093
    if(trailing_ones<total_coeff) {
4094
        int level_code, mask;
4095
        int suffix_length = total_coeff > 10 && trailing_ones < 3;
4096
        int prefix= get_level_prefix(gb);
4097

    
4098
        //first coefficient has suffix_length equal to 0 or 1
4099
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4100
            if(suffix_length)
4101
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4102
            else
4103
                level_code= (prefix<<suffix_length); //part
4104
        }else if(prefix==14){
4105
            if(suffix_length)
4106
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4107
            else
4108
                level_code= prefix + get_bits(gb, 4); //part
4109
        }else{
4110
            level_code= (15<<suffix_length) + get_bits(gb, prefix-3); //part
4111
            if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
4112
            if(prefix>=16)
4113
                level_code += (1<<(prefix-3))-4096;
4114
        }
4115

    
4116
        if(trailing_ones < 3) level_code += 2;
4117

    
4118
        suffix_length = 1;
4119
        if(level_code > 5)
4120
            suffix_length++;
4121
        mask= -(level_code&1);
4122
        level[trailing_ones]= (((2+level_code)>>1) ^ mask) - mask;
4123

    
4124
        //remaining coefficients have suffix_length > 0
4125
        for(i=trailing_ones+1;i<total_coeff;i++) {
4126
            static const int suffix_limit[7] = {0,5,11,23,47,95,INT_MAX };
4127
            prefix = get_level_prefix(gb);
4128
            if(prefix<15){
4129
                level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
4130
            }else{
4131
                level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
4132
                if(prefix>=16)
4133
                    level_code += (1<<(prefix-3))-4096;
4134
            }
4135
            mask= -(level_code&1);
4136
            level[i]= (((2+level_code)>>1) ^ mask) - mask;
4137
            if(level_code > suffix_limit[suffix_length])
4138
                suffix_length++;
4139
        }
4140
    }
4141

    
4142
    if(total_coeff == max_coeff)
4143
        zeros_left=0;
4144
    else{
4145
        if(n == CHROMA_DC_BLOCK_INDEX)
4146
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4147
        else
4148
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4149
    }
4150

    
4151
    coeff_num = zeros_left + total_coeff - 1;
4152
    j = scantable[coeff_num];
4153
    if(n > 24){
4154
        block[j] = level[0];
4155
        for(i=1;i<total_coeff;i++) {
4156
            if(zeros_left <= 0)
4157
                run_before = 0;
4158
            else if(zeros_left < 7){
4159
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4160
            }else{
4161
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4162
            }
4163
            zeros_left -= run_before;
4164
            coeff_num -= 1 + run_before;
4165
            j= scantable[ coeff_num ];
4166

    
4167
            block[j]= level[i];
4168
        }
4169
    }else{
4170
        block[j] = (level[0] * qmul[j] + 32)>>6;
4171
        for(i=1;i<total_coeff;i++) {
4172
            if(zeros_left <= 0)
4173
                run_before = 0;
4174
            else if(zeros_left < 7){
4175
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4176
            }else{
4177
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4178
            }
4179
            zeros_left -= run_before;
4180
            coeff_num -= 1 + run_before;
4181
            j= scantable[ coeff_num ];
4182

    
4183
            block[j]= (level[i] * qmul[j] + 32)>>6;
4184
        }
4185
    }
4186

    
4187
    if(zeros_left<0){
4188
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4189
        return -1;
4190
    }
4191

    
4192
    return 0;
4193
}
4194

    
4195
static void predict_field_decoding_flag(H264Context *h){
4196
    MpegEncContext * const s = &h->s;
4197
    const int mb_xy= h->mb_xy;
4198
    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
4199
                ? s->current_picture.mb_type[mb_xy-1]
4200
                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
4201
                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
4202
                : 0;
4203
    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4204
}
4205

    
4206
/**
4207
 * decodes a P_SKIP or B_SKIP macroblock
4208
 */
4209
static void decode_mb_skip(H264Context *h){
4210
    MpegEncContext * const s = &h->s;
4211
    const int mb_xy= h->mb_xy;
4212
    int mb_type=0;
4213

    
4214
    memset(h->non_zero_count[mb_xy], 0, 16);
4215
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4216

    
4217
    if(MB_FIELD)
4218
        mb_type|= MB_TYPE_INTERLACED;
4219

    
4220
    if( h->slice_type_nos == FF_B_TYPE )
4221
    {
4222
        // just for fill_caches. pred_direct_motion will set the real mb_type
4223
        mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4224

    
4225
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4226
        pred_direct_motion(h, &mb_type);
4227
        mb_type|= MB_TYPE_SKIP;
4228
    }
4229
    else
4230
    {
4231
        int mx, my;
4232
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4233

    
4234
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4235
        pred_pskip_motion(h, &mx, &my);
4236
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4237
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4238
    }
4239

    
4240
    write_back_motion(h, mb_type);
4241
    s->current_picture.mb_type[mb_xy]= mb_type;
4242
    s->current_picture.qscale_table[mb_xy]= s->qscale;
4243
    h->slice_table[ mb_xy ]= h->slice_num;
4244
    h->prev_mb_skipped= 1;
4245
}
4246

    
4247
/**
4248
 * decodes a macroblock
4249
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4250
 */
4251
static int decode_mb_cavlc(H264Context *h){
4252
    MpegEncContext * const s = &h->s;
4253
    int mb_xy;
4254
    int partition_count;
4255
    unsigned int mb_type, cbp;
4256
    int dct8x8_allowed= h->pps.transform_8x8_mode;
4257

    
4258
    mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4259

    
4260
    tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4261
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4262
                down the code */
4263
    if(h->slice_type_nos != FF_I_TYPE){
4264
        if(s->mb_skip_run==-1)
4265
            s->mb_skip_run= get_ue_golomb(&s->gb);
4266

    
4267
        if (s->mb_skip_run--) {
4268
            if(FRAME_MBAFF && (s->mb_y&1) == 0){
4269
                if(s->mb_skip_run==0)
4270
                    h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4271
                else
4272
                    predict_field_decoding_flag(h);
4273
            }
4274
            decode_mb_skip(h);
4275
            return 0;
4276
        }
4277
    }
4278
    if(FRAME_MBAFF){
4279
        if( (s->mb_y&1) == 0 )
4280
            h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4281
    }
4282

    
4283
    h->prev_mb_skipped= 0;
4284

    
4285
    mb_type= get_ue_golomb(&s->gb);
4286
    if(h->slice_type_nos == FF_B_TYPE){
4287
        if(mb_type < 23){
4288
            partition_count= b_mb_type_info[mb_type].partition_count;
4289
            mb_type=         b_mb_type_info[mb_type].type;
4290
        }else{
4291
            mb_type -= 23;
4292
            goto decode_intra_mb;
4293
        }
4294
    }else if(h->slice_type_nos == FF_P_TYPE){
4295
        if(mb_type < 5){
4296
            partition_count= p_mb_type_info[mb_type].partition_count;
4297
            mb_type=         p_mb_type_info[mb_type].type;
4298
        }else{
4299
            mb_type -= 5;
4300
            goto decode_intra_mb;
4301
        }
4302
    }else{
4303
       assert(h->slice_type_nos == FF_I_TYPE);
4304
        if(h->slice_type == FF_SI_TYPE && mb_type)
4305
            mb_type--;
4306
decode_intra_mb:
4307
        if(mb_type > 25){
4308
            av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice too large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
4309
            return -1;
4310
        }
4311
        partition_count=0;
4312
        cbp= i_mb_type_info[mb_type].cbp;
4313
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4314
        mb_type= i_mb_type_info[mb_type].type;
4315
    }
4316

    
4317
    if(MB_FIELD)
4318
        mb_type |= MB_TYPE_INTERLACED;
4319

    
4320
    h->slice_table[ mb_xy ]= h->slice_num;
4321

    
4322
    if(IS_INTRA_PCM(mb_type)){
4323
        unsigned int x;
4324

    
4325
        // We assume these blocks are very rare so we do not optimize it.
4326
        align_get_bits(&s->gb);
4327

    
4328
        // The pixels are stored in the same order as levels in h->mb array.
4329
        for(x=0; x < (CHROMA ? 384 : 256); x++){
4330
            ((uint8_t*)h->mb)[x]= get_bits(&s->gb, 8);
4331
        }
4332

    
4333
        // In deblocking, the quantizer is 0
4334
        s->current_picture.qscale_table[mb_xy]= 0;
4335
        // All coeffs are present
4336
        memset(h->non_zero_count[mb_xy], 16, 16);
4337

    
4338
        s->current_picture.mb_type[mb_xy]= mb_type;
4339
        return 0;
4340
    }
4341

    
4342
    if(MB_MBAFF){
4343
        h->ref_count[0] <<= 1;
4344
        h->ref_count[1] <<= 1;
4345
    }
4346

    
4347
    fill_caches(h, mb_type, 0);
4348

    
4349
    //mb_pred
4350
    if(IS_INTRA(mb_type)){
4351
        int pred_mode;
4352
//            init_top_left_availability(h);
4353
        if(IS_INTRA4x4(mb_type)){
4354
            int i;
4355
            int di = 1;
4356
            if(dct8x8_allowed && get_bits1(&s->gb)){
4357
                mb_type |= MB_TYPE_8x8DCT;
4358
                di = 4;
4359
            }
4360

    
4361
//                fill_intra4x4_pred_table(h);
4362
            for(i=0; i<16; i+=di){
4363
                int mode= pred_intra_mode(h, i);
4364

    
4365
                if(!get_bits1(&s->gb)){
4366
                    const int rem_mode= get_bits(&s->gb, 3);
4367
                    mode = rem_mode + (rem_mode >= mode);
4368
                }
4369

    
4370
                if(di==4)
4371
                    fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
4372
                else
4373
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4374
            }
4375
            write_back_intra_pred_mode(h);
4376
            if( check_intra4x4_pred_mode(h) < 0)
4377
                return -1;
4378
        }else{
4379
            h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4380
            if(h->intra16x16_pred_mode < 0)
4381
                return -1;
4382
        }
4383
        if(CHROMA){
4384
            pred_mode= check_intra_pred_mode(h, get_ue_golomb(&s->gb));
4385
            if(pred_mode < 0)
4386
                return -1;
4387
            h->chroma_pred_mode= pred_mode;
4388
        }
4389
    }else if(partition_count==4){
4390
        int i, j, sub_partition_count[4], list, ref[2][4];
4391

    
4392
        if(h->slice_type_nos == FF_B_TYPE){
4393
            for(i=0; i<4; i++){
4394
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4395
                if(h->sub_mb_type[i] >=13){
4396
                    av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
4397
                    return -1;
4398
                }
4399
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4400
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4401
            }
4402
            if(   IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4403
               || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
4404
                pred_direct_motion(h, &mb_type);
4405
                h->ref_cache[0][scan8[4]] =
4406
                h->ref_cache[1][scan8[4]] =
4407
                h->ref_cache[0][scan8[12]] =
4408
                h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
4409
            }
4410
        }else{
4411
            assert(h->slice_type_nos == FF_P_TYPE); //FIXME SP correct ?
4412
            for(i=0; i<4; i++){
4413
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4414
                if(h->sub_mb_type[i] >=4){
4415
                    av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
4416
                    return -1;
4417
                }
4418
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4419
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4420
            }
4421
        }
4422

    
4423
        for(list=0; list<h->list_count; list++){
4424
            int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4425
            for(i=0; i<4; i++){
4426
                if(IS_DIRECT(h->sub_mb_type[i])) continue;
4427
                if(IS_DIR(h->sub_mb_type[i], 0, list)){
4428
                    unsigned int tmp = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4429
                    if(tmp>=ref_count){
4430
                        av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", tmp);
4431
                        return -1;
4432
                    }
4433
                    ref[list][i]= tmp;
4434
                }else{
4435
                 //FIXME
4436
                    ref[list][i] = -1;
4437
                }
4438
            }
4439
        }
4440

    
4441
        if(dct8x8_allowed)
4442
            dct8x8_allowed = get_dct8x8_allowed(h);
4443

    
4444
        for(list=0; list<h->list_count; list++){
4445
            for(i=0; i<4; i++){
4446
                if(IS_DIRECT(h->sub_mb_type[i])) {
4447
                    h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ];
4448
                    continue;
4449
                }
4450
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4451
                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4452

    
4453
                if(IS_DIR(h->sub_mb_type[i], 0, list)){
4454
                    const int sub_mb_type= h->sub_mb_type[i];
4455
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4456
                    for(j=0; j<sub_partition_count[i]; j++){
4457
                        int mx, my;
4458
                        const int index= 4*i + block_width*j;
4459
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4460
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4461
                        mx += get_se_golomb(&s->gb);
4462