Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ e08715d3

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
                left_block= left_block_options[2 - bottom];
159
            }
160
        }
161
    }
162

    
163
    h->top_mb_xy = top_xy;
164
    h->left_mb_xy[0] = left_xy[0];
165
    h->left_mb_xy[1] = left_xy[1];
166
    if(for_deblock){
167
        topleft_type = 0;
168
        topright_type = 0;
169
        top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
170
        left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
171
        left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
172

    
173
        if(MB_MBAFF && !IS_INTRA(mb_type)){
174
            int list;
175
            for(list=0; list<h->list_count; list++){
176
                //These values where changed for ease of performing MC, we need to change them back
177
                //FIXME maybe we can make MC and loop filter use the same values or prevent
178
                //the MC code from changing ref_cache and rather use a temporary array.
179
                if(USES_LIST(mb_type,list)){
180
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
181
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
182
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
183
                    ref += h->b8_stride;
184
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
185
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
186
                }
187
            }
188
        }
189
    }else{
190
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
191
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
192
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
193
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
194
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
195

    
196
    if(IS_INTRA(mb_type)){
197
        int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
198
        h->topleft_samples_available=
199
        h->top_samples_available=
200
        h->left_samples_available= 0xFFFF;
201
        h->topright_samples_available= 0xEEEA;
202

    
203
        if(!(top_type & type_mask)){
204
            h->topleft_samples_available= 0xB3FF;
205
            h->top_samples_available= 0x33FF;
206
            h->topright_samples_available= 0x26EA;
207
        }
208
        if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
209
            if(IS_INTERLACED(mb_type)){
210
                if(!(left_type[0] & type_mask)){
211
                    h->topleft_samples_available&= 0xDFFF;
212
                    h->left_samples_available&= 0x5FFF;
213
                }
214
                if(!(left_type[1] & type_mask)){
215
                    h->topleft_samples_available&= 0xFF5F;
216
                    h->left_samples_available&= 0xFF5F;
217
                }
218
            }else{
219
                int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
220
                                ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
221
                assert(left_xy[0] == left_xy[1]);
222
                if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
223
                    h->topleft_samples_available&= 0xDF5F;
224
                    h->left_samples_available&= 0x5F5F;
225
                }
226
            }
227
        }else{
228
            if(!(left_type[0] & type_mask)){
229
                h->topleft_samples_available&= 0xDF5F;
230
                h->left_samples_available&= 0x5F5F;
231
            }
232
        }
233

    
234
        if(!(topleft_type & type_mask))
235
            h->topleft_samples_available&= 0x7FFF;
236

    
237
        if(!(topright_type & type_mask))
238
            h->topright_samples_available&= 0xFBFF;
239

    
240
        if(IS_INTRA4x4(mb_type)){
241
            if(IS_INTRA4x4(top_type)){
242
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
243
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
244
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
245
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
246
            }else{
247
                int pred;
248
                if(!(top_type & type_mask))
249
                    pred= -1;
250
                else{
251
                    pred= 2;
252
                }
253
                h->intra4x4_pred_mode_cache[4+8*0]=
254
                h->intra4x4_pred_mode_cache[5+8*0]=
255
                h->intra4x4_pred_mode_cache[6+8*0]=
256
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
257
            }
258
            for(i=0; i<2; i++){
259
                if(IS_INTRA4x4(left_type[i])){
260
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
261
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
262
                }else{
263
                    int pred;
264
                    if(!(left_type[i] & type_mask))
265
                        pred= -1;
266
                    else{
267
                        pred= 2;
268
                    }
269
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
270
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
271
                }
272
            }
273
        }
274
    }
275
    }
276

    
277

    
278
/*
279
0 . T T. T T T T
280
1 L . .L . . . .
281
2 L . .L . . . .
282
3 . T TL . . . .
283
4 L . .L . . . .
284
5 L . .. . . . .
285
*/
286
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
287
    if(top_type){
288
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
289
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
290
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
291
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
292

    
293
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
294
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
295

    
296
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
297
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
298

    
299
    }else{
300
        h->non_zero_count_cache[4+8*0]=
301
        h->non_zero_count_cache[5+8*0]=
302
        h->non_zero_count_cache[6+8*0]=
303
        h->non_zero_count_cache[7+8*0]=
304

    
305
        h->non_zero_count_cache[1+8*0]=
306
        h->non_zero_count_cache[2+8*0]=
307

    
308
        h->non_zero_count_cache[1+8*3]=
309
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
310

    
311
    }
312

    
313
    for (i=0; i<2; i++) {
314
        if(left_type[i]){
315
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
316
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
317
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
318
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
319
        }else{
320
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
321
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
322
            h->non_zero_count_cache[0+8*1 +   8*i]=
323
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
324
        }
325
    }
326

    
327
    if( h->pps.cabac ) {
328
        // top_cbp
329
        if(top_type) {
330
            h->top_cbp = h->cbp_table[top_xy];
331
        } else if(IS_INTRA(mb_type)) {
332
            h->top_cbp = 0x1C0;
333
        } else {
334
            h->top_cbp = 0;
335
        }
336
        // left_cbp
337
        if (left_type[0]) {
338
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
339
        } else if(IS_INTRA(mb_type)) {
340
            h->left_cbp = 0x1C0;
341
        } else {
342
            h->left_cbp = 0;
343
        }
344
        if (left_type[0]) {
345
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
346
        }
347
        if (left_type[1]) {
348
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
349
        }
350
    }
351

    
352
#if 1
353
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
354
        int list;
355
        for(list=0; list<h->list_count; list++){
356
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
357
                /*if(!h->mv_cache_clean[list]){
358
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
359
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
360
                    h->mv_cache_clean[list]= 1;
361
                }*/
362
                continue;
363
            }
364
            h->mv_cache_clean[list]= 0;
365

    
366
            if(USES_LIST(top_type, list)){
367
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
368
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
369
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
370
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
371
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
372
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
373
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
374
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
375
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
376
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
377
            }else{
378
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
379
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
380
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
381
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
382
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
383
            }
384

    
385
            for(i=0; i<2; i++){
386
                int cache_idx = scan8[0] - 1 + i*2*8;
387
                if(USES_LIST(left_type[i], list)){
388
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
389
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
390
                    *(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]];
391
                    *(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]];
392
                    h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
393
                    h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
394
                }else{
395
                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
396
                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
397
                    h->ref_cache[list][cache_idx  ]=
398
                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
399
                }
400
            }
401

    
402
            if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
403
                continue;
404

    
405
            if(USES_LIST(topleft_type, list)){
406
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
407
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
408
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
409
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
410
            }else{
411
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
412
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
413
            }
414

    
415
            if(USES_LIST(topright_type, list)){
416
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
417
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
418
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
419
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
420
            }else{
421
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
422
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
423
            }
424

    
425
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
426
                continue;
427

    
428
            h->ref_cache[list][scan8[5 ]+1] =
429
            h->ref_cache[list][scan8[7 ]+1] =
430
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
431
            h->ref_cache[list][scan8[4 ]] =
432
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
433
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
434
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
435
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
436
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
437
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
438

    
439
            if( h->pps.cabac ) {
440
                /* XXX beurk, Load mvd */
441
                if(USES_LIST(top_type, list)){
442
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
443
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
444
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
445
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
446
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
447
                }else{
448
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
449
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
450
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
451
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
452
                }
453
                if(USES_LIST(left_type[0], list)){
454
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
455
                    *(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]];
456
                    *(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]];
457
                }else{
458
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
459
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
460
                }
461
                if(USES_LIST(left_type[1], list)){
462
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
463
                    *(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]];
464
                    *(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]];
465
                }else{
466
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
467
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
468
                }
469
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
470
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
471
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
472
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
473
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
474

    
475
                if(h->slice_type_nos == FF_B_TYPE){
476
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
477

    
478
                    if(IS_DIRECT(top_type)){
479
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
480
                    }else if(IS_8X8(top_type)){
481
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
482
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
483
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
484
                    }else{
485
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
486
                    }
487

    
488
                    if(IS_DIRECT(left_type[0]))
489
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
490
                    else if(IS_8X8(left_type[0]))
491
                        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)];
492
                    else
493
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
494

    
495
                    if(IS_DIRECT(left_type[1]))
496
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
497
                    else if(IS_8X8(left_type[1]))
498
                        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)];
499
                    else
500
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
501
                }
502
            }
503

    
504
            if(FRAME_MBAFF){
505
#define MAP_MVS\
506
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
507
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
508
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
509
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
510
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
511
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
512
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
513
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
514
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
515
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
516
                if(MB_FIELD){
517
#define MAP_F2F(idx, mb_type)\
518
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
519
                        h->ref_cache[list][idx] <<= 1;\
520
                        h->mv_cache[list][idx][1] /= 2;\
521
                        h->mvd_cache[list][idx][1] /= 2;\
522
                    }
523
                    MAP_MVS
524
#undef MAP_F2F
525
                }else{
526
#define MAP_F2F(idx, mb_type)\
527
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
528
                        h->ref_cache[list][idx] >>= 1;\
529
                        h->mv_cache[list][idx][1] <<= 1;\
530
                        h->mvd_cache[list][idx][1] <<= 1;\
531
                    }
532
                    MAP_MVS
533
#undef MAP_F2F
534
                }
535
            }
536
        }
537
    }
538
#endif
539

    
540
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
541
}
542

    
543
static inline void write_back_intra_pred_mode(H264Context *h){
544
    const int mb_xy= h->mb_xy;
545

    
546
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
547
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
548
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
549
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
550
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
551
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
552
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
553
}
554

    
555
/**
556
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
557
 */
558
static inline int check_intra4x4_pred_mode(H264Context *h){
559
    MpegEncContext * const s = &h->s;
560
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
561
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
562
    int i;
563

    
564
    if(!(h->top_samples_available&0x8000)){
565
        for(i=0; i<4; i++){
566
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
567
            if(status<0){
568
                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);
569
                return -1;
570
            } else if(status){
571
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
572
            }
573
        }
574
    }
575

    
576
    if((h->left_samples_available&0x8888)!=0x8888){
577
        static const int mask[4]={0x8000,0x2000,0x80,0x20};
578
        for(i=0; i<4; i++){
579
            if(!(h->left_samples_available&mask[i])){
580
                int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
581
                if(status<0){
582
                    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);
583
                    return -1;
584
                } else if(status){
585
                    h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
586
                }
587
            }
588
        }
589
    }
590

    
591
    return 0;
592
} //FIXME cleanup like next
593

    
594
/**
595
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
596
 */
597
static inline int check_intra_pred_mode(H264Context *h, int mode){
598
    MpegEncContext * const s = &h->s;
599
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
600
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
601

    
602
    if(mode > 6U) {
603
        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);
604
        return -1;
605
    }
606

    
607
    if(!(h->top_samples_available&0x8000)){
608
        mode= top[ mode ];
609
        if(mode<0){
610
            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);
611
            return -1;
612
        }
613
    }
614

    
615
    if((h->left_samples_available&0x8080) != 0x8080){
616
        mode= left[ mode ];
617
        if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
618
            mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
619
        }
620
        if(mode<0){
621
            av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
622
            return -1;
623
        }
624
    }
625

    
626
    return mode;
627
}
628

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

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

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

    
644
static inline void write_back_non_zero_count(H264Context *h){
645
    const int mb_xy= h->mb_xy;
646

    
647
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
648
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
649
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
650
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
651
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
652
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
653
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
654

    
655
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
656
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
657
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
658

    
659
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
660
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
661
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
662
}
663

    
664
/**
665
 * gets the predicted number of non-zero coefficients.
666
 * @param n block index
667
 */
668
static inline int pred_non_zero_count(H264Context *h, int n){
669
    const int index8= scan8[n];
670
    const int left= h->non_zero_count_cache[index8 - 1];
671
    const int top = h->non_zero_count_cache[index8 - 8];
672
    int i= left + top;
673

    
674
    if(i<64) i= (i+1)>>1;
675

    
676
    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
677

    
678
    return i&31;
679
}
680

    
681
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
682
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
683
    MpegEncContext *s = &h->s;
684

    
685
    /* there is no consistent mapping of mvs to neighboring locations that will
686
     * make mbaff happy, so we can't move all this logic to fill_caches */
687
    if(FRAME_MBAFF){
688
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
689
        const int16_t *mv;
690
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
691
        *C = h->mv_cache[list][scan8[0]-2];
692

    
693
        if(!MB_FIELD
694
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
695
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
696
            if(IS_INTERLACED(mb_types[topright_xy])){
697
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
698
                const int x4 = X4, y4 = Y4;\
699
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
700
                if(!USES_LIST(mb_type,list))\
701
                    return LIST_NOT_USED;\
702
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
703
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
704
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
705
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
706

    
707
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
708
            }
709
        }
710
        if(topright_ref == PART_NOT_AVAILABLE
711
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
712
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
713
            if(!MB_FIELD
714
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
715
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
716
            }
717
            if(MB_FIELD
718
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
719
               && i >= scan8[0]+8){
720
                // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
721
                SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
722
            }
723
        }
724
#undef SET_DIAG_MV
725
    }
726

    
727
    if(topright_ref != PART_NOT_AVAILABLE){
728
        *C= h->mv_cache[list][ i - 8 + part_width ];
729
        return topright_ref;
730
    }else{
731
        tprintf(s->avctx, "topright MV not available\n");
732

    
733
        *C= h->mv_cache[list][ i - 8 - 1 ];
734
        return h->ref_cache[list][ i - 8 - 1 ];
735
    }
736
}
737

    
738
/**
739
 * gets the predicted MV.
740
 * @param n the block index
741
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
742
 * @param mx the x component of the predicted motion vector
743
 * @param my the y component of the predicted motion vector
744
 */
745
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
746
    const int index8= scan8[n];
747
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
748
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
749
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
750
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
751
    const int16_t * C;
752
    int diagonal_ref, match_count;
753

    
754
    assert(part_width==1 || part_width==2 || part_width==4);
755

    
756
/* mv_cache
757
  B . . A T T T T
758
  U . . L . . , .
759
  U . . L . . . .
760
  U . . L . . , .
761
  . . . L . . . .
762
*/
763

    
764
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
765
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
766
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
767
    if(match_count > 1){ //most common
768
        *mx= mid_pred(A[0], B[0], C[0]);
769
        *my= mid_pred(A[1], B[1], C[1]);
770
    }else if(match_count==1){
771
        if(left_ref==ref){
772
            *mx= A[0];
773
            *my= A[1];
774
        }else if(top_ref==ref){
775
            *mx= B[0];
776
            *my= B[1];
777
        }else{
778
            *mx= C[0];
779
            *my= C[1];
780
        }
781
    }else{
782
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
783
            *mx= A[0];
784
            *my= A[1];
785
        }else{
786
            *mx= mid_pred(A[0], B[0], C[0]);
787
            *my= mid_pred(A[1], B[1], C[1]);
788
        }
789
    }
790

    
791
    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);
792
}
793

    
794
/**
795
 * gets the directionally predicted 16x8 MV.
796
 * @param n the block index
797
 * @param mx the x component of the predicted motion vector
798
 * @param my the y component of the predicted motion vector
799
 */
800
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
801
    if(n==0){
802
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
803
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
804

    
805
        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);
806

    
807
        if(top_ref == ref){
808
            *mx= B[0];
809
            *my= B[1];
810
            return;
811
        }
812
    }else{
813
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
814
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
815

    
816
        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);
817

    
818
        if(left_ref == ref){
819
            *mx= A[0];
820
            *my= A[1];
821
            return;
822
        }
823
    }
824

    
825
    //RARE
826
    pred_motion(h, n, 4, list, ref, mx, my);
827
}
828

    
829
/**
830
 * gets the directionally predicted 8x16 MV.
831
 * @param n the block index
832
 * @param mx the x component of the predicted motion vector
833
 * @param my the y component of the predicted motion vector
834
 */
835
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
836
    if(n==0){
837
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
838
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
839

    
840
        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);
841

    
842
        if(left_ref == ref){
843
            *mx= A[0];
844
            *my= A[1];
845
            return;
846
        }
847
    }else{
848
        const int16_t * C;
849
        int diagonal_ref;
850

    
851
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
852

    
853
        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);
854

    
855
        if(diagonal_ref == ref){
856
            *mx= C[0];
857
            *my= C[1];
858
            return;
859
        }
860
    }
861

    
862
    //RARE
863
    pred_motion(h, n, 2, list, ref, mx, my);
864
}
865

    
866
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
867
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
868
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
869

    
870
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
871

    
872
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
873
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
874
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
875

    
876
        *mx = *my = 0;
877
        return;
878
    }
879

    
880
    pred_motion(h, 0, 4, 0, 0, mx, my);
881

    
882
    return;
883
}
884

    
885
static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
886
    int poc0 = h->ref_list[0][i].poc;
887
    int td = av_clip(poc1 - poc0, -128, 127);
888
    if(td == 0 || h->ref_list[0][i].long_ref){
889
        return 256;
890
    }else{
891
        int tb = av_clip(poc - poc0, -128, 127);
892
        int tx = (16384 + (FFABS(td) >> 1)) / td;
893
        return av_clip((tb*tx + 32) >> 6, -1024, 1023);
894
    }
895
}
896

    
897
static inline void direct_dist_scale_factor(H264Context * const h){
898
    MpegEncContext * const s = &h->s;
899
    const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
900
    const int poc1 = h->ref_list[1][0].poc;
901
    int i, field;
902
    for(field=0; field<2; field++){
903
        const int poc  = h->s.current_picture_ptr->field_poc[field];
904
        const int poc1 = h->ref_list[1][0].field_poc[field];
905
        for(i=0; i < 2*h->ref_count[0]; i++)
906
            h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
907
    }
908

    
909
    for(i=0; i<h->ref_count[0]; i++){
910
        h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
911
    }
912
}
913

    
914
static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
915
    MpegEncContext * const s = &h->s;
916
    Picture * const ref1 = &h->ref_list[1][0];
917
    int j, old_ref, rfield;
918
    int start= mbafi ? 16                      : 0;
919
    int end  = mbafi ? 16+2*h->ref_count[list] : h->ref_count[list];
920
    int interl= mbafi || s->picture_structure != PICT_FRAME;
921

    
922
    /* bogus; fills in for missing frames */
923
    memset(map[list], 0, sizeof(map[list]));
924

    
925
    for(rfield=0; rfield<2; rfield++){
926
        for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
927
            int poc = ref1->ref_poc[colfield][list][old_ref];
928

    
929
            if     (!interl)
930
                poc |= 3;
931
            else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
932
                poc= (poc&~3) + rfield + 1;
933

    
934
            for(j=start; j<end; j++){
935
                if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
936
                    int cur_ref= mbafi ? (j-16)^field : j;
937
                    map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
938
                    if(rfield == field)
939
                        map[list][old_ref] = cur_ref;
940
                    break;
941
                }
942
            }
943
        }
944
    }
945
}
946

    
947
static inline void direct_ref_list_init(H264Context * const h){
948
    MpegEncContext * const s = &h->s;
949
    Picture * const ref1 = &h->ref_list[1][0];
950
    Picture * const cur = s->current_picture_ptr;
951
    int list, j, field;
952
    int sidx= (s->picture_structure&1)^1;
953
    int ref1sidx= (ref1->reference&1)^1;
954

    
955
    for(list=0; list<2; list++){
956
        cur->ref_count[sidx][list] = h->ref_count[list];
957
        for(j=0; j<h->ref_count[list]; j++)
958
            cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
959
    }
960

    
961
    if(s->picture_structure == PICT_FRAME){
962
        memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
963
        memcpy(cur->ref_poc  [1], cur->ref_poc  [0], sizeof(cur->ref_poc  [0]));
964
    }
965

    
966
    cur->mbaff= FRAME_MBAFF;
967

    
968
    if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
969
        return;
970

    
971
    for(list=0; list<2; list++){
972
        fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
973
        for(field=0; field<2; field++)
974
            fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
975
    }
976
}
977

    
978
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
979
    MpegEncContext * const s = &h->s;
980
    int b8_stride = h->b8_stride;
981
    int b4_stride = h->b_stride;
982
    int mb_xy = h->mb_xy;
983
    int mb_type_col[2];
984
    const int16_t (*l1mv0)[2], (*l1mv1)[2];
985
    const int8_t *l1ref0, *l1ref1;
986
    const int is_b8x8 = IS_8X8(*mb_type);
987
    unsigned int sub_mb_type;
988
    int i8, i4;
989

    
990
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
991

    
992
    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
993
        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
994
            int cur_poc = s->current_picture_ptr->poc;
995
            int *col_poc = h->ref_list[1]->field_poc;
996
            int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
997
            mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
998
            b8_stride = 0;
999
        }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity
1000
            int fieldoff= 2*(h->ref_list[1][0].reference)-3;
1001
            mb_xy += s->mb_stride*fieldoff;
1002
        }
1003
        goto single_col;
1004
    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
1005
        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
1006
            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
1007
            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
1008
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
1009
            b8_stride *= 3;
1010
            b4_stride *= 6;
1011
            //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag
1012
            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
1013
                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
1014
                && !is_b8x8){
1015
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1016
                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
1017
            }else{
1018
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1019
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1020
            }
1021
        }else{                                           //     AFR/FR    -> AFR/FR
1022
single_col:
1023
            mb_type_col[0] =
1024
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
1025
            if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
1026
                /* FIXME save sub mb types from previous frames (or derive from MVs)
1027
                * so we know exactly what block size to use */
1028
                sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1029
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1030
            }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
1031
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1032
                *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1033
            }else{
1034
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1035
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1036
            }
1037
        }
1038
    }
1039

    
1040
    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
1041
    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
1042
    l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
1043
    l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
1044
    if(!b8_stride){
1045
        if(s->mb_y&1){
1046
            l1ref0 += h->b8_stride;
1047
            l1ref1 += h->b8_stride;
1048
            l1mv0  +=  2*b4_stride;
1049
            l1mv1  +=  2*b4_stride;
1050
        }
1051
    }
1052

    
1053
    if(h->direct_spatial_mv_pred){
1054
        int ref[2];
1055
        int mv[2][2];
1056
        int list;
1057

    
1058
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1059

    
1060
        /* ref = min(neighbors) */
1061
        for(list=0; list<2; list++){
1062
            int refa = h->ref_cache[list][scan8[0] - 1];
1063
            int refb = h->ref_cache[list][scan8[0] - 8];
1064
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1065
            if(refc == PART_NOT_AVAILABLE)
1066
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1067
            ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
1068
            if(ref[list] < 0)
1069
                ref[list] = -1;
1070
        }
1071

    
1072
        if(ref[0] < 0 && ref[1] < 0){
1073
            ref[0] = ref[1] = 0;
1074
            mv[0][0] = mv[0][1] =
1075
            mv[1][0] = mv[1][1] = 0;
1076
        }else{
1077
            for(list=0; list<2; list++){
1078
                if(ref[list] >= 0)
1079
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1080
                else
1081
                    mv[list][0] = mv[list][1] = 0;
1082
            }
1083
        }
1084

    
1085
        if(ref[1] < 0){
1086
            if(!is_b8x8)
1087
                *mb_type &= ~MB_TYPE_L1;
1088
            sub_mb_type &= ~MB_TYPE_L1;
1089
        }else if(ref[0] < 0){
1090
            if(!is_b8x8)
1091
                *mb_type &= ~MB_TYPE_L0;
1092
            sub_mb_type &= ~MB_TYPE_L0;
1093
        }
1094

    
1095
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1096
            for(i8=0; i8<4; i8++){
1097
                int x8 = i8&1;
1098
                int y8 = i8>>1;
1099
                int xy8 = x8+y8*b8_stride;
1100
                int xy4 = 3*x8+y8*b4_stride;
1101
                int a=0, b=0;
1102

    
1103
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1104
                    continue;
1105
                h->sub_mb_type[i8] = sub_mb_type;
1106

    
1107
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1108
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1109
                if(!IS_INTRA(mb_type_col[y8])
1110
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1111
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1112
                    if(ref[0] > 0)
1113
                        a= pack16to32(mv[0][0],mv[0][1]);
1114
                    if(ref[1] > 0)
1115
                        b= pack16to32(mv[1][0],mv[1][1]);
1116
                }else{
1117
                    a= pack16to32(mv[0][0],mv[0][1]);
1118
                    b= pack16to32(mv[1][0],mv[1][1]);
1119
                }
1120
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1121
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1122
            }
1123
        }else if(IS_16X16(*mb_type)){
1124
            int a=0, b=0;
1125

    
1126
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1127
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1128
            if(!IS_INTRA(mb_type_col[0])
1129
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1130
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1131
                       && (h->x264_build>33 || !h->x264_build)))){
1132
                if(ref[0] > 0)
1133
                    a= pack16to32(mv[0][0],mv[0][1]);
1134
                if(ref[1] > 0)
1135
                    b= pack16to32(mv[1][0],mv[1][1]);
1136
            }else{
1137
                a= pack16to32(mv[0][0],mv[0][1]);
1138
                b= pack16to32(mv[1][0],mv[1][1]);
1139
            }
1140
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1141
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1142
        }else{
1143
            for(i8=0; i8<4; i8++){
1144
                const int x8 = i8&1;
1145
                const int y8 = i8>>1;
1146

    
1147
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1148
                    continue;
1149
                h->sub_mb_type[i8] = sub_mb_type;
1150

    
1151
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1152
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1153
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1154
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1155

    
1156
                /* col_zero_flag */
1157
                if(!IS_INTRA(mb_type_col[0]) && (   l1ref0[x8 + y8*b8_stride] == 0
1158
                                              || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
1159
                                                  && (h->x264_build>33 || !h->x264_build)))){
1160
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
1161
                    if(IS_SUB_8X8(sub_mb_type)){
1162
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1163
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1164
                            if(ref[0] == 0)
1165
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1166
                            if(ref[1] == 0)
1167
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1168
                        }
1169
                    }else
1170
                    for(i4=0; i4<4; i4++){
1171
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1172
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1173
                            if(ref[0] == 0)
1174
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1175
                            if(ref[1] == 0)
1176
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1177
                        }
1178
                    }
1179
                }
1180
            }
1181
        }
1182
    }else{ /* direct temporal mv pred */
1183
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1184
        const int *dist_scale_factor = h->dist_scale_factor;
1185
        int ref_offset= 0;
1186

    
1187
        if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
1188
            map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
1189
            map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
1190
            dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];
1191
        }
1192
        if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
1193
            ref_offset += 16;
1194

    
1195
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1196
            /* FIXME assumes direct_8x8_inference == 1 */
1197
            int y_shift  = 2*!IS_INTERLACED(*mb_type);
1198

    
1199
            for(i8=0; i8<4; i8++){
1200
                const int x8 = i8&1;
1201
                const int y8 = i8>>1;
1202
                int ref0, scale;
1203
                const int16_t (*l1mv)[2]= l1mv0;
1204

    
1205
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1206
                    continue;
1207
                h->sub_mb_type[i8] = sub_mb_type;
1208

    
1209
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1210
                if(IS_INTRA(mb_type_col[y8])){
1211
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1212
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1213
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1214
                    continue;
1215
                }
1216

    
1217
                ref0 = l1ref0[x8 + y8*b8_stride];
1218
                if(ref0 >= 0)
1219
                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
1220
                else{
1221
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1222
                    l1mv= l1mv1;
1223
                }
1224
                scale = dist_scale_factor[ref0];
1225
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1226

    
1227
                {
1228
                    const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
1229
                    int my_col = (mv_col[1]<<y_shift)/2;
1230
                    int mx = (scale * mv_col[0] + 128) >> 8;
1231
                    int my = (scale * my_col + 128) >> 8;
1232
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1233
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1234
                }
1235
            }
1236
            return;
1237
        }
1238

    
1239
        /* one-to-one mv scaling */
1240

    
1241
        if(IS_16X16(*mb_type)){
1242
            int ref, mv0, mv1;
1243

    
1244
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1245
            if(IS_INTRA(mb_type_col[0])){
1246
                ref=mv0=mv1=0;
1247
            }else{
1248
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
1249
                                                : map_col_to_list0[1][l1ref1[0] + ref_offset];
1250
                const int scale = dist_scale_factor[ref0];
1251
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1252
                int mv_l0[2];
1253
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1254
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1255
                ref= ref0;
1256
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
1257
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1258
            }
1259
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1260
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1261
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1262
        }else{
1263
            for(i8=0; i8<4; i8++){
1264
                const int x8 = i8&1;
1265
                const int y8 = i8>>1;
1266
                int ref0, scale;
1267
                const int16_t (*l1mv)[2]= l1mv0;
1268

    
1269
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1270
                    continue;
1271
                h->sub_mb_type[i8] = sub_mb_type;
1272
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1273
                if(IS_INTRA(mb_type_col[0])){
1274
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1275
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1276
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1277
                    continue;
1278
                }
1279

    
1280
                ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset;
1281
                if(ref0 >= 0)
1282
                    ref0 = map_col_to_list0[0][ref0];
1283
                else{
1284
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1285
                    l1mv= l1mv1;
1286
                }
1287
                scale = dist_scale_factor[ref0];
1288

    
1289
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1290
                if(IS_SUB_8X8(sub_mb_type)){
1291
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1292
                    int mx = (scale * mv_col[0] + 128) >> 8;
1293
                    int my = (scale * mv_col[1] + 128) >> 8;
1294
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1295
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1296
                }else
1297
                for(i4=0; i4<4; i4++){
1298
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1299
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1300
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1301
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1302
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1303
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1304
                }
1305
            }
1306
        }
1307
    }
1308
}
1309

    
1310
static inline void write_back_motion(H264Context *h, int mb_type){
1311
    MpegEncContext * const s = &h->s;
1312
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1313
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1314
    int list;
1315

    
1316
    if(!USES_LIST(mb_type, 0))
1317
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1318

    
1319
    for(list=0; list<h->list_count; list++){
1320
        int y;
1321
        if(!USES_LIST(mb_type, list))
1322
            continue;
1323

    
1324
        for(y=0; y<4; y++){
1325
            *(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];
1326
            *(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];
1327
        }
1328
        if( h->pps.cabac ) {
1329
            if(IS_SKIP(mb_type))
1330
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1331
            else
1332
            for(y=0; y<4; y++){
1333
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1334
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1335
            }
1336
        }
1337

    
1338
        {
1339
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1340
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1341
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1342
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1343
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1344
        }
1345
    }
1346

    
1347
    if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
1348
        if(IS_8X8(mb_type)){
1349
            uint8_t *direct_table = &h->direct_table[b8_xy];
1350
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1351
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1352
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1353
        }
1354
    }
1355
}
1356

    
1357
/**
1358
 * Decodes a network abstraction layer unit.
1359
 * @param consumed is the number of bytes used as input
1360
 * @param length is the length of the array
1361
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1362
 * @returns decoded bytes, might be src+1 if no escapes
1363
 */
1364
static const uint8_t *decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1365
    int i, si, di;
1366
    uint8_t *dst;
1367
    int bufidx;
1368

    
1369
//    src[0]&0x80;                //forbidden bit
1370
    h->nal_ref_idc= src[0]>>5;
1371
    h->nal_unit_type= src[0]&0x1F;
1372

    
1373
    src++; length--;
1374
#if 0
1375
    for(i=0; i<length; i++)
1376
        printf("%2X ", src[i]);
1377
#endif
1378

    
1379
#ifdef HAVE_FAST_UNALIGNED
1380
# ifdef HAVE_FAST_64BIT
1381
#   define RS 7
1382
    for(i=0; i+1<length; i+=9){
1383
        if(!((~*(uint64_t*)(src+i) & (*(uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
1384
# else
1385
#   define RS 3
1386
    for(i=0; i+1<length; i+=5){
1387
        if(!((~*(uint32_t*)(src+i) & (*(uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
1388
# endif
1389
            continue;
1390
        if(i>0 && !src[i]) i--;
1391
        while(src[i]) i++;
1392
#else
1393
#   define RS 0
1394
    for(i=0; i+1<length; i+=2){
1395
        if(src[i]) continue;
1396
        if(i>0 && src[i-1]==0) i--;
1397
#endif
1398
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1399
            if(src[i+2]!=3){
1400
                /* startcode, so we must be past the end */
1401
                length=i;
1402
            }
1403
            break;
1404
        }
1405
    }
1406

    
1407
    if(i>=length-1){ //no escaped 0
1408
        *dst_length= length;
1409
        *consumed= length+1; //+1 for the header
1410
        return src;
1411
    }
1412

    
1413
    bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1414
    h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
1415
    dst= h->rbsp_buffer[bufidx];
1416

    
1417
    if (dst == NULL){
1418
        return NULL;
1419
    }
1420

    
1421
//printf("decoding esc\n");
1422
    si=di=0;
1423
    while(si<length){
1424
        //remove escapes (very rare 1:2^22)
1425
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1426
            if(src[si+2]==3){ //escape
1427
                dst[di++]= 0;
1428
                dst[di++]= 0;
1429
                si+=3;
1430
                continue;
1431
            }else //next start code
1432
                break;
1433
        }
1434

    
1435
        dst[di++]= src[si++];
1436
    }
1437

    
1438
    memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
1439

    
1440
    *dst_length= di;
1441
    *consumed= si + 1;//+1 for the header
1442
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1443
    return dst;
1444
}
1445

    
1446
/**
1447
 * identifies the exact end of the bitstream
1448
 * @return the length of the trailing, or 0 if damaged
1449
 */
1450
static int decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1451
    int v= *src;
1452
    int r;
1453

    
1454
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1455

    
1456
    for(r=1; r<9; r++){
1457
        if(v&1) return r;
1458
        v>>=1;
1459
    }
1460
    return 0;
1461
}
1462

    
1463
/**
1464
 * IDCT transforms the 16 dc values and dequantizes them.
1465
 * @param qp quantization parameter
1466
 */
1467
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1468
#define stride 16
1469
    int i;
1470
    int temp[16]; //FIXME check if this is a good idea
1471
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1472
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1473

    
1474
//memset(block, 64, 2*256);
1475
//return;
1476
    for(i=0; i<4; i++){
1477
        const int offset= y_offset[i];
1478
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1479
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1480
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1481
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1482

    
1483
        temp[4*i+0]= z0+z3;
1484
        temp[4*i+1]= z1+z2;
1485
        temp[4*i+2]= z1-z2;
1486
        temp[4*i+3]= z0-z3;
1487
    }
1488

    
1489
    for(i=0; i<4; i++){
1490
        const int offset= x_offset[i];
1491
        const int z0= temp[4*0+i] + temp[4*2+i];
1492
        const int z1= temp[4*0+i] - temp[4*2+i];
1493
        const int z2= temp[4*1+i] - temp[4*3+i];
1494
        const int z3= temp[4*1+i] + temp[4*3+i];
1495

    
1496
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
1497
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1498
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1499
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1500
    }
1501
}
1502

    
1503
#if 0
1504
/**
1505
 * DCT transforms the 16 dc values.
1506
 * @param qp quantization parameter ??? FIXME
1507
 */
1508
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1509
//    const int qmul= dequant_coeff[qp][0];
1510
    int i;
1511
    int temp[16]; //FIXME check if this is a good idea
1512
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1513
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1514

1515
    for(i=0; i<4; i++){
1516
        const int offset= y_offset[i];
1517
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1518
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1519
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1520
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1521

1522
        temp[4*i+0]= z0+z3;
1523
        temp[4*i+1]= z1+z2;
1524
        temp[4*i+2]= z1-z2;
1525
        temp[4*i+3]= z0-z3;
1526
    }
1527

1528
    for(i=0; i<4; i++){
1529
        const int offset= x_offset[i];
1530
        const int z0= temp[4*0+i] + temp[4*2+i];
1531
        const int z1= temp[4*0+i] - temp[4*2+i];
1532
        const int z2= temp[4*1+i] - temp[4*3+i];
1533
        const int z3= temp[4*1+i] + temp[4*3+i];
1534

1535
        block[stride*0 +offset]= (z0 + z3)>>1;
1536
        block[stride*2 +offset]= (z1 + z2)>>1;
1537
        block[stride*8 +offset]= (z1 - z2)>>1;
1538
        block[stride*10+offset]= (z0 - z3)>>1;
1539
    }
1540
}
1541
#endif
1542

    
1543
#undef xStride
1544
#undef stride
1545

    
1546
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1547
    const int stride= 16*2;
1548
    const int xStride= 16;
1549
    int a,b,c,d,e;
1550

    
1551
    a= block[stride*0 + xStride*0];
1552
    b= block[stride*0 + xStride*1];
1553
    c= block[stride*1 + xStride*0];
1554
    d= block[stride*1 + xStride*1];
1555

    
1556
    e= a-b;
1557
    a= a+b;
1558
    b= c-d;
1559
    c= c+d;
1560

    
1561
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1562
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1563
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1564
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1565
}
1566

    
1567
#if 0
1568
static void chroma_dc_dct_c(DCTELEM *block){
1569
    const int stride= 16*2;
1570
    const int xStride= 16;
1571
    int a,b,c,d,e;
1572

1573
    a= block[stride*0 + xStride*0];
1574
    b= block[stride*0 + xStride*1];
1575
    c= block[stride*1 + xStride*0];
1576
    d= block[stride*1 + xStride*1];
1577

1578
    e= a-b;
1579
    a= a+b;
1580
    b= c-d;
1581
    c= c+d;
1582

1583
    block[stride*0 + xStride*0]= (a+c);
1584
    block[stride*0 + xStride*1]= (e+b);
1585
    block[stride*1 + xStride*0]= (a-c);
1586
    block[stride*1 + xStride*1]= (e-b);
1587
}
1588
#endif
1589

    
1590
/**
1591
 * gets the chroma qp.
1592
 */
1593
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1594
    return h->pps.chroma_qp_table[t][qscale];
1595
}
1596

    
1597
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1598
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1599
                           int src_x_offset, int src_y_offset,
1600
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1601
    MpegEncContext * const s = &h->s;
1602
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1603
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1604
    const int luma_xy= (mx&3) + ((my&3)<<2);
1605
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1606
    uint8_t * src_cb, * src_cr;
1607
    int extra_width= h->emu_edge_width;
1608
    int extra_height= h->emu_edge_height;
1609
    int emu=0;
1610
    const int full_mx= mx>>2;
1611
    const int full_my= my>>2;
1612
    const int pic_width  = 16*s->mb_width;
1613
    const int pic_height = 16*s->mb_height >> MB_FIELD;
1614

    
1615
    if(mx&7) extra_width -= 3;
1616
    if(my&7) extra_height -= 3;
1617

    
1618
    if(   full_mx < 0-extra_width
1619
       || full_my < 0-extra_height
1620
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
1621
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
1622
        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);
1623
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1624
        emu=1;
1625
    }
1626

    
1627
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1628
    if(!square){
1629
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1630
    }
1631

    
1632
    if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1633

    
1634
    if(MB_FIELD){
1635
        // chroma offset when predicting from a field of opposite parity
1636
        my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1637
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1638
    }
1639
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1640
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1641

    
1642
    if(emu){
1643
        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);
1644
            src_cb= s->edge_emu_buffer;
1645
    }
1646
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1647

    
1648
    if(emu){
1649
        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);
1650
            src_cr= s->edge_emu_buffer;
1651
    }
1652
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1653
}
1654

    
1655
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1656
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1657
                           int x_offset, int y_offset,
1658
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1659
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1660
                           int list0, int list1){
1661
    MpegEncContext * const s = &h->s;
1662
    qpel_mc_func *qpix_op=  qpix_put;
1663
    h264_chroma_mc_func chroma_op= chroma_put;
1664

    
1665
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1666
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1667
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1668
    x_offset += 8*s->mb_x;
1669
    y_offset += 8*(s->mb_y >> MB_FIELD);
1670

    
1671
    if(list0){
1672
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1673
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1674
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1675
                           qpix_op, chroma_op);
1676

    
1677
        qpix_op=  qpix_avg;
1678
        chroma_op= chroma_avg;
1679
    }
1680

    
1681
    if(list1){
1682
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1683
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1684
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1685
                           qpix_op, chroma_op);
1686
    }
1687
}
1688

    
1689
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1690
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1691
                           int x_offset, int y_offset,
1692
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1693
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1694
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1695
                           int list0, int list1){
1696
    MpegEncContext * const s = &h->s;
1697

    
1698
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1699
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1700
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1701
    x_offset += 8*s->mb_x;
1702
    y_offset += 8*(s->mb_y >> MB_FIELD);
1703

    
1704
    if(list0 && list1){
1705
        /* don't optimize for luma-only case, since B-frames usually
1706
         * use implicit weights => chroma too. */
1707
        uint8_t *tmp_cb = s->obmc_scratchpad;
1708
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1709
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1710
        int refn0 = h->ref_cache[0][ scan8[n] ];
1711
        int refn1 = h->ref_cache[1][ scan8[n] ];
1712

    
1713
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1714
                    dest_y, dest_cb, dest_cr,
1715
                    x_offset, y_offset, qpix_put, chroma_put);
1716
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1717
                    tmp_y, tmp_cb, tmp_cr,
1718
                    x_offset, y_offset, qpix_put, chroma_put);
1719

    
1720
        if(h->use_weight == 2){
1721
            int weight0 = h->implicit_weight[refn0][refn1];
1722
            int weight1 = 64 - weight0;
1723
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
1724
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1725
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1726
        }else{
1727
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1728
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1729
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1730
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1731
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1732
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1733
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1734
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1735
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1736
        }
1737
    }else{
1738
        int list = list1 ? 1 : 0;
1739
        int refn = h->ref_cache[list][ scan8[n] ];
1740
        Picture *ref= &h->ref_list[list][refn];
1741
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1742
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
1743
                    qpix_put, chroma_put);
1744

    
1745
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1746
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
1747
        if(h->use_weight_chroma){
1748
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1749
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1750
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1751
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1752
        }
1753
    }
1754
}
1755

    
1756
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1757
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1758
                           int x_offset, int y_offset,
1759
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1760
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1761
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1762
                           int list0, int list1){
1763
    if((h->use_weight==2 && list0 && list1
1764
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1765
       || h->use_weight==1)
1766
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1767
                         x_offset, y_offset, qpix_put, chroma_put,
1768
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1769
    else
1770
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1771
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1772
}
1773

    
1774
static inline void prefetch_motion(H264Context *h, int list){
1775
    /* fetch pixels for estimated mv 4 macroblocks ahead
1776
     * optimized for 64byte cache lines */
1777
    MpegEncContext * const s = &h->s;
1778
    const int refn = h->ref_cache[list][scan8[0]];
1779
    if(refn >= 0){
1780
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1781
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1782
        uint8_t **src= h->ref_list[list][refn].data;
1783
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1784
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
1785
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1786
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1787
    }
1788
}
1789

    
1790
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1791
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1792
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1793
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1794
    MpegEncContext * const s = &h->s;
1795
    const int mb_xy= h->mb_xy;
1796
    const int mb_type= s->current_picture.mb_type[mb_xy];
1797

    
1798
    assert(IS_INTER(mb_type));
1799

    
1800
    prefetch_motion(h, 0);
1801

    
1802
    if(IS_16X16(mb_type)){
1803
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1804
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1805
                &weight_op[0], &weight_avg[0],
1806
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1807
    }else if(IS_16X8(mb_type)){
1808
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1809
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1810
                &weight_op[1], &weight_avg[1],
1811
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1812
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1813
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1814
                &weight_op[1], &weight_avg[1],
1815
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1816
    }else if(IS_8X16(mb_type)){
1817
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1818
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1819
                &weight_op[2], &weight_avg[2],
1820
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1821
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1822
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1823
                &weight_op[2], &weight_avg[2],
1824
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1825
    }else{
1826
        int i;
1827

    
1828
        assert(IS_8X8(mb_type));
1829

    
1830
        for(i=0; i<4; i++){
1831
            const int sub_mb_type= h->sub_mb_type[i];
1832
            const int n= 4*i;
1833
            int x_offset= (i&1)<<2;
1834
            int y_offset= (i&2)<<1;
1835

    
1836
            if(IS_SUB_8X8(sub_mb_type)){
1837
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1838
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1839
                    &weight_op[3], &weight_avg[3],
1840
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1841
            }else if(IS_SUB_8X4(sub_mb_type)){
1842
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1843
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1844
                    &weight_op[4], &weight_avg[4],
1845
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1846
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1847
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1848
                    &weight_op[4], &weight_avg[4],
1849
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1850
            }else if(IS_SUB_4X8(sub_mb_type)){
1851
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1852
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1853
                    &weight_op[5], &weight_avg[5],
1854
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1855
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1856
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1857
                    &weight_op[5], &weight_avg[5],
1858
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1859
            }else{
1860
                int j;
1861
                assert(IS_SUB_4X4(sub_mb_type));
1862
                for(j=0; j<4; j++){
1863
                    int sub_x_offset= x_offset + 2*(j&1);
1864
                    int sub_y_offset= y_offset +   (j&2);
1865
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1866
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1867
                        &weight_op[6], &weight_avg[6],
1868
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1869
                }
1870
            }
1871
        }
1872
    }
1873

    
1874
    prefetch_motion(h, 1);
1875
}
1876

    
1877
static av_cold void decode_init_vlc(void){
1878
    static int done = 0;
1879

    
1880
    if (!done) {
1881
        int i;
1882
        int offset;
1883
        done = 1;
1884

    
1885
        chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
1886
        chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
1887
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1888
                 &chroma_dc_coeff_token_len [0], 1, 1,
1889
                 &chroma_dc_coeff_token_bits[0], 1, 1,
1890
                 INIT_VLC_USE_NEW_STATIC);
1891

    
1892
        offset = 0;
1893
        for(i=0; i<4; i++){
1894
            coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
1895
            coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
1896
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1897
                     &coeff_token_len [i][0], 1, 1,
1898
                     &coeff_token_bits[i][0], 1, 1,
1899
                     INIT_VLC_USE_NEW_STATIC);
1900
            offset += coeff_token_vlc_tables_size[i];
1901
        }
1902
        /*
1903
         * This is a one time safety check to make sure that
1904
         * the packed static coeff_token_vlc table sizes
1905
         * were initialized correctly.
1906
         */
1907
        assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
1908

    
1909
        for(i=0; i<3; i++){
1910
            chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
1911
            chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
1912
            init_vlc(&chroma_dc_total_zeros_vlc[i],
1913
                     CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1914
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
1915
                     &chroma_dc_total_zeros_bits[i][0], 1, 1,
1916
                     INIT_VLC_USE_NEW_STATIC);
1917
        }
1918
        for(i=0; i<15; i++){
1919
            total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
1920
            total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
1921
            init_vlc(&total_zeros_vlc[i],
1922
                     TOTAL_ZEROS_VLC_BITS, 16,
1923
                     &total_zeros_len [i][0], 1, 1,
1924
                     &total_zeros_bits[i][0], 1, 1,
1925
                     INIT_VLC_USE_NEW_STATIC);
1926
        }
1927

    
1928
        for(i=0; i<6; i++){
1929
            run_vlc[i].table = run_vlc_tables[i];
1930
            run_vlc[i].table_allocated = run_vlc_tables_size;
1931
            init_vlc(&run_vlc[i],
1932
                     RUN_VLC_BITS, 7,
1933
                     &run_len [i][0], 1, 1,
1934
                     &run_bits[i][0], 1, 1,
1935
                     INIT_VLC_USE_NEW_STATIC);
1936
        }
1937
        run7_vlc.table = run7_vlc_table,
1938
        run7_vlc.table_allocated = run7_vlc_table_size;
1939
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1940
                 &run_len [6][0], 1, 1,
1941
                 &run_bits[6][0], 1, 1,
1942
                 INIT_VLC_USE_NEW_STATIC);
1943
    }
1944
}
1945

    
1946
static void free_tables(H264Context *h){
1947
    int i;
1948
    H264Context *hx;
1949
    av_freep(&h->intra4x4_pred_mode);
1950
    av_freep(&h->chroma_pred_mode_table);
1951
    av_freep(&h->cbp_table);
1952
    av_freep(&h->mvd_table[0]);
1953
    av_freep(&h->mvd_table[1]);
1954
    av_freep(&h->direct_table);
1955
    av_freep(&h->non_zero_count);
1956
    av_freep(&h->slice_table_base);
1957
    h->slice_table= NULL;
1958

    
1959
    av_freep(&h->mb2b_xy);
1960
    av_freep(&h->mb2b8_xy);
1961

    
1962
    for(i = 0; i < h->s.avctx->thread_count; i++) {
1963
        hx = h->thread_context[i];
1964
        if(!hx) continue;
1965
        av_freep(&hx->top_borders[1]);
1966
        av_freep(&hx->top_borders[0]);
1967
        av_freep(&hx->s.obmc_scratchpad);
1968
    }
1969
}
1970

    
1971
static void init_dequant8_coeff_table(H264Context *h){
1972
    int i,q,x;
1973
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
1974
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
1975
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
1976

    
1977
    for(i=0; i<2; i++ ){
1978
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
1979
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
1980
            break;
1981
        }
1982

    
1983
        for(q=0; q<52; q++){
1984
            int shift = div6[q];
1985
            int idx = rem6[q];
1986
            for(x=0; x<64; x++)
1987
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
1988
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
1989
                    h->pps.scaling_matrix8[i][x]) << shift;
1990
        }
1991
    }
1992
}
1993

    
1994
static void init_dequant4_coeff_table(H264Context *h){
1995
    int i,j,q,x;
1996
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
1997
    for(i=0; i<6; i++ ){
1998
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
1999
        for(j=0; j<i; j++){
2000
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2001
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2002
                break;
2003
            }
2004
        }
2005
        if(j<i)
2006
            continue;
2007

    
2008
        for(q=0; q<52; q++){
2009
            int shift = div6[q] + 2;
2010
            int idx = rem6[q];
2011
            for(x=0; x<16; x++)
2012
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2013
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2014
                    h->pps.scaling_matrix4[i][x]) << shift;
2015
        }
2016
    }
2017
}
2018

    
2019
static void init_dequant_tables(H264Context *h){
2020
    int i,x;
2021
    init_dequant4_coeff_table(h);
2022
    if(h->pps.transform_8x8_mode)
2023
        init_dequant8_coeff_table(h);
2024
    if(h->sps.transform_bypass){
2025
        for(i=0; i<6; i++)
2026
            for(x=0; x<16; x++)
2027
                h->dequant4_coeff[i][0][x] = 1<<6;
2028
        if(h->pps.transform_8x8_mode)
2029
            for(i=0; i<2; i++)
2030
                for(x=0; x<64; x++)
2031
                    h->dequant8_coeff[i][0][x] = 1<<6;
2032
    }
2033
}
2034

    
2035

    
2036
/**
2037
 * allocates tables.
2038
 * needs width/height
2039
 */
2040
static int alloc_tables(H264Context *h){
2041
    MpegEncContext * const s = &h->s;
2042
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2043
    int x,y;
2044

    
2045
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2046

    
2047
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2048
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base))
2049
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2050

    
2051
    CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2052
    CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2053
    CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2054
    CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2055

    
2056
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
2057
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2058

    
2059
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2060
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2061
    for(y=0; y<s->mb_height; y++){
2062
        for(x=0; x<s->mb_width; x++){
2063
            const int mb_xy= x + y*s->mb_stride;
2064
            const int b_xy = 4*x + 4*y*h->b_stride;
2065
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2066

    
2067
            h->mb2b_xy [mb_xy]= b_xy;
2068
            h->mb2b8_xy[mb_xy]= b8_xy;
2069
        }
2070
    }
2071

    
2072
    s->obmc_scratchpad = NULL;
2073

    
2074
    if(!h->dequant4_coeff[0])
2075
        init_dequant_tables(h);
2076

    
2077
    return 0;
2078
fail:
2079
    free_tables(h);
2080
    return -1;
2081
}
2082

    
2083
/**
2084
 * Mimic alloc_tables(), but for every context thread.
2085
 */
2086
static void clone_tables(H264Context *dst, H264Context *src){
2087
    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
2088
    dst->non_zero_count           = src->non_zero_count;
2089
    dst->slice_table              = src->slice_table;
2090
    dst->cbp_table                = src->cbp_table;
2091
    dst->mb2b_xy                  = src->mb2b_xy;
2092
    dst->mb2b8_xy                 = src->mb2b8_xy;
2093
    dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
2094
    dst->mvd_table[0]             = src->mvd_table[0];
2095
    dst->mvd_table[1]             = src->mvd_table[1];
2096
    dst->direct_table             = src->direct_table;
2097

    
2098
    dst->s.obmc_scratchpad = NULL;
2099
    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2100
}
2101

    
2102
/**
2103
 * Init context
2104
 * Allocate buffers which are not shared amongst multiple threads.
2105
 */
2106
static int context_init(H264Context *h){
2107
    CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2108
    CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2109

    
2110
    return 0;
2111
fail:
2112
    return -1; // free_tables will clean up for us
2113
}
2114

    
2115
static av_cold void common_init(H264Context *h){
2116
    MpegEncContext * const s = &h->s;
2117

    
2118
    s->width = s->avctx->width;
2119
    s->height = s->avctx->height;
2120
    s->codec_id= s->avctx->codec->id;
2121

    
2122
    ff_h264_pred_init(&h->hpc, s->codec_id);
2123

    
2124
    h->dequant_coeff_pps= -1;
2125
    s->unrestricted_mv=1;
2126
    s->decode=1; //FIXME
2127

    
2128
    dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
2129

    
2130
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2131
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2132
}
2133

    
2134
static av_cold int decode_init(AVCodecContext *avctx){
2135
    H264Context *h= avctx->priv_data;
2136
    MpegEncContext * const s = &h->s;
2137

    
2138
    MPV_decode_defaults(s);
2139

    
2140
    s->avctx = avctx;
2141
    common_init(h);
2142

    
2143
    s->out_format = FMT_H264;
2144
    s->workaround_bugs= avctx->workaround_bugs;
2145

    
2146
    // set defaults
2147
//    s->decode_mb= ff_h263_decode_mb;
2148
    s->quarter_sample = 1;
2149
    s->low_delay= 1;
2150

    
2151
    if(avctx->codec_id == CODEC_ID_SVQ3)
2152
        avctx->pix_fmt= PIX_FMT_YUVJ420P;
2153
    else
2154
        avctx->pix_fmt= PIX_FMT_YUV420P;
2155

    
2156
    decode_init_vlc();
2157

    
2158
    if(avctx->extradata_size > 0 && avctx->extradata &&
2159
       *(char *)avctx->extradata == 1){
2160
        h->is_avc = 1;
2161
        h->got_avcC = 0;
2162
    } else {
2163
        h->is_avc = 0;
2164
    }
2165

    
2166
    h->thread_context[0] = h;
2167
    h->outputed_poc = INT_MIN;
2168
    h->prev_poc_msb= 1<<16;
2169
    return 0;
2170
}
2171

    
2172
static int frame_start(H264Context *h){
2173
    MpegEncContext * const s = &h->s;
2174
    int i;
2175

    
2176
    if(MPV_frame_start(s, s->avctx) < 0)
2177
        return -1;
2178
    ff_er_frame_start(s);
2179
    /*
2180
     * MPV_frame_start uses pict_type to derive key_frame.
2181
     * This is incorrect for H.264; IDR markings must be used.
2182
     * Zero here; IDR markings per slice in frame or fields are ORed in later.
2183
     * See decode_nal_units().
2184
     */
2185
    s->current_picture_ptr->key_frame= 0;
2186

    
2187
    assert(s->linesize && s->uvlinesize);
2188

    
2189
    for(i=0; i<16; i++){
2190
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2191
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2192
    }
2193
    for(i=0; i<4; i++){
2194
        h->block_offset[16+i]=
2195
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2196
        h->block_offset[24+16+i]=
2197
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2198
    }
2199

    
2200
    /* can't be in alloc_tables because linesize isn't known there.
2201
     * FIXME: redo bipred weight to not require extra buffer? */
2202
    for(i = 0; i < s->avctx->thread_count; i++)
2203
        if(!h->thread_context[i]->s.obmc_scratchpad)
2204
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2205

    
2206
    /* some macroblocks will be accessed before they're available */
2207
    if(FRAME_MBAFF || s->avctx->thread_count > 1)
2208
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
2209

    
2210
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2211

    
2212
    // We mark the current picture as non-reference after allocating it, so
2213
    // that if we break out due to an error it can be released automatically
2214
    // in the next MPV_frame_start().
2215
    // SVQ3 as well as most other codecs have only last/next/current and thus
2216
    // get released even with set reference, besides SVQ3 and others do not
2217
    // mark frames as reference later "naturally".
2218
    if(s->codec_id != CODEC_ID_SVQ3)
2219
        s->current_picture_ptr->reference= 0;
2220

    
2221
    s->current_picture_ptr->field_poc[0]=
2222
    s->current_picture_ptr->field_poc[1]= INT_MAX;
2223
    assert(s->current_picture_ptr->long_ref==0);
2224

    
2225
    return 0;
2226
}
2227

    
2228
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){
2229
    MpegEncContext * const s = &h->s;
2230
    int i;
2231
    int step    = 1;
2232
    int offset  = 1;
2233
    int uvoffset= 1;
2234
    int top_idx = 1;
2235
    int skiplast= 0;
2236

    
2237
    src_y  -=   linesize;
2238
    src_cb -= uvlinesize;
2239
    src_cr -= uvlinesize;
2240

    
2241
    if(!simple && FRAME_MBAFF){
2242
        if(s->mb_y&1){
2243
            offset  = MB_MBAFF ? 1 : 17;
2244
            uvoffset= MB_MBAFF ? 1 : 9;
2245
            if(!MB_MBAFF){
2246
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y +  15*linesize);
2247
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
2248
                if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2249
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
2250
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
2251
                }
2252
            }
2253
        }else{
2254
            if(!MB_MBAFF){
2255
                h->left_border[0]= h->top_borders[0][s->mb_x][15];
2256
                if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2257
                    h->left_border[34   ]= h->top_borders[0][s->mb_x][16+7  ];
2258
                    h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
2259
                }
2260
                skiplast= 1;
2261
            }
2262
            offset  =
2263
            uvoffset=
2264
            top_idx = MB_MBAFF ? 0 : 1;
2265
        }
2266
        step= MB_MBAFF ? 2 : 1;
2267
    }
2268

    
2269
    // There are two lines saved, the line above the the top macroblock of a pair,
2270
    // and the line above the bottom macroblock
2271
    h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
2272
    for(i=1; i<17 - skiplast; i++){
2273
        h->left_border[offset+i*step]= src_y[15+i*  linesize];
2274
    }
2275

    
2276
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2277
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2278

    
2279
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2280
        h->left_border[uvoffset+34   ]= h->top_borders[top_idx][s->mb_x][16+7];
2281
        h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
2282
        for(i=1; i<9 - skiplast; i++){
2283
            h->left_border[uvoffset+34   +i*step]= src_cb[7+i*uvlinesize];
2284
            h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
2285
        }
2286
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2287
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2288
    }
2289
}
2290

    
2291
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){
2292
    MpegEncContext * const s = &h->s;
2293
    int temp8, i;
2294
    uint64_t temp64;
2295
    int deblock_left;
2296
    int deblock_top;
2297
    int mb_xy;
2298
    int step    = 1;
2299
    int offset  = 1;
2300
    int uvoffset= 1;
2301
    int top_idx = 1;
2302

    
2303
    if(!simple && FRAME_MBAFF){
2304
        if(s->mb_y&1){
2305
            offset  = MB_MBAFF ? 1 : 17;
2306
            uvoffset= MB_MBAFF ? 1 : 9;
2307
        }else{
2308
            offset  =
2309
            uvoffset=
2310
            top_idx = MB_MBAFF ? 0 : 1;
2311
        }
2312
        step= MB_MBAFF ? 2 : 1;
2313
    }
2314

    
2315
    if(h->deblocking_filter == 2) {
2316
        mb_xy = h->mb_xy;
2317
        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2318
        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2319
    } else {
2320
        deblock_left = (s->mb_x > 0);
2321
        deblock_top =  (s->mb_y > !!MB_FIELD);
2322
    }
2323

    
2324
    src_y  -=   linesize + 1;
2325
    src_cb -= uvlinesize + 1;
2326
    src_cr -= uvlinesize + 1;
2327

    
2328
#define XCHG(a,b,t,xchg)\
2329
t= a;\
2330
if(xchg)\
2331
    a= b;\
2332
b= t;
2333

    
2334
    if(deblock_left){
2335
        for(i = !deblock_top; i<16; i++){
2336
            XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, xchg);
2337
        }
2338
        XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, 1);
2339
    }
2340

    
2341
    if(deblock_top){
2342
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2343
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2344
        if(s->mb_x+1 < s->mb_width){
2345
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2346
        }
2347
    }
2348

    
2349
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2350
        if(deblock_left){
2351
            for(i = !deblock_top; i<8; i++){
2352
                XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, xchg);
2353
                XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
2354
            }
2355
            XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, 1);
2356
            XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
2357
        }
2358
        if(deblock_top){
2359
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2360
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2361
        }
2362
    }
2363
}
2364

    
2365
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2366
    MpegEncContext * const s = &h->s;
2367
    const int mb_x= s->mb_x;
2368
    const int mb_y= s->mb_y;
2369
    const int mb_xy= h->mb_xy;
2370
    const int mb_type= s->current_picture.mb_type[mb_xy];
2371
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2372
    int linesize, uvlinesize /*dct_offset*/;
2373
    int i;
2374
    int *block_offset = &h->block_offset[0];
2375
    const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2376
    const int is_h264 = simple || s->codec_id == CODEC_ID_H264;
2377
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2378
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2379

    
2380
    dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2381
    dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2382
    dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2383

    
2384
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2385
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2386

    
2387
    if (!simple && MB_FIELD) {
2388
        linesize   = h->mb_linesize   = s->linesize * 2;
2389
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2390
        block_offset = &h->block_offset[24];
2391
        if(mb_y&1){ //FIXME move out of this function?
2392
            dest_y -= s->linesize*15;
2393
            dest_cb-= s->uvlinesize*7;
2394
            dest_cr-= s->uvlinesize*7;
2395
        }
2396
        if(FRAME_MBAFF) {
2397
            int list;
2398
            for(list=0; list<h->list_count; list++){
2399
                if(!USES_LIST(mb_type, list))
2400
                    continue;
2401
                if(IS_16X16(mb_type)){
2402
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
2403
                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2404
                }else{
2405
                    for(i=0; i<16; i+=4){
2406
                        int ref = h->ref_cache[list][scan8[i]];
2407
                        if(ref >= 0)
2408
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2409
                    }
2410
                }
2411
            }
2412
        }
2413
    } else {
2414
        linesize   = h->mb_linesize   = s->linesize;
2415
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2416
//        dct_offset = s->linesize * 16;
2417
    }
2418

    
2419
    if (!simple && IS_INTRA_PCM(mb_type)) {
2420
        for (i=0; i<16; i++) {
2421
            memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
2422
        }
2423
        for (i=0; i<8; i++) {
2424
            memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
2425
            memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
2426
        }
2427
    } else {
2428
        if(IS_INTRA(mb_type)){
2429
            if(h->deblocking_filter)
2430
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2431

    
2432
            if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2433
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2434
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2435
            }
2436

    
2437
            if(IS_INTRA4x4(mb_type)){
2438
                if(simple || !s->encoding){
2439
                    if(IS_8x8DCT(mb_type)){
2440
                        if(transform_bypass){
2441
                            idct_dc_add =
2442
                            idct_add    = s->dsp.add_pixels8;
2443
                        }else{
2444
                            idct_dc_add = s->dsp.h264_idct8_dc_add;
2445
                            idct_add    = s->dsp.h264_idct8_add;
2446
                        }
2447
                        for(i=0; i<16; i+=4){
2448
                            uint8_t * const ptr= dest_y + block_offset[i];
2449
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2450
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2451
                                h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
2452
                            }else{
2453
                                const int nnz = h->non_zero_count_cache[ scan8[i] ];
2454
                                h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2455
                                                            (h->topright_samples_available<<i)&0x4000, linesize);
2456
                                if(nnz){
2457
                                    if(nnz == 1 && h->mb[i*16])
2458
                                        idct_dc_add(ptr, h->mb + i*16, linesize);
2459
                                    else
2460
                                        idct_add   (ptr, h->mb + i*16, linesize);
2461
                                }
2462
                            }
2463
                        }
2464
                    }else{
2465
                        if(transform_bypass){
2466
                            idct_dc_add =
2467
                            idct_add    = s->dsp.add_pixels4;
2468
                        }else{
2469
                            idct_dc_add = s->dsp.h264_idct_dc_add;
2470
                            idct_add    = s->dsp.h264_idct_add;
2471
                        }
2472
                        for(i=0; i<16; i++){
2473
                            uint8_t * const ptr= dest_y + block_offset[i];
2474
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2475

    
2476
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2477
                                h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
2478
                            }else{
2479
                                uint8_t *topright;
2480
                                int nnz, tr;
2481
                                if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2482
                                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2483
                                    assert(mb_y || linesize <= block_offset[i]);
2484
                                    if(!topright_avail){
2485
                                        tr= ptr[3 - linesize]*0x01010101;
2486
                                        topright= (uint8_t*) &tr;
2487
                                    }else
2488
                                        topright= ptr + 4 - linesize;
2489
                                }else
2490
                                    topright= NULL;
2491

    
2492
                                h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2493
                                nnz = h->non_zero_count_cache[ scan8[i] ];
2494
                                if(nnz){
2495
                                    if(is_h264){
2496
                                        if(nnz == 1 && h->mb[i*16])
2497
                                            idct_dc_add(ptr, h->mb + i*16, linesize);
2498
                                        else
2499
                                            idct_add   (ptr, h->mb + i*16, linesize);
2500
                                    }else
2501
                                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2502
                                }
2503
                            }
2504
                        }
2505
                    }
2506
                }
2507
            }else{
2508
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2509
                if(is_h264){
2510
                    if(!transform_bypass)
2511
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
2512
                }else
2513
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2514
            }
2515
            if(h->deblocking_filter)
2516
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2517
        }else if(is_h264){
2518
            hl_motion(h, dest_y, dest_cb, dest_cr,
2519
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2520
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2521
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2522
        }
2523

    
2524

    
2525
        if(!IS_INTRA4x4(mb_type)){
2526
            if(is_h264){
2527
                if(IS_INTRA16x16(mb_type)){
2528
                    if(transform_bypass){
2529
                        if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
2530
                            h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
2531
                        }else{
2532
                            for(i=0; i<16; i++){
2533
                                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2534
                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
2535
                            }
2536
                        }
2537
                    }else{
2538
                         s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2539
                    }
2540
                }else if(h->cbp&15){
2541
                    if(transform_bypass){
2542
                        const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2543
                        idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2544
                        for(i=0; i<16; i+=di){
2545
                            if(h->non_zero_count_cache[ scan8[i] ]){
2546
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2547
                            }
2548
                        }
2549
                    }else{
2550
                        if(IS_8x8DCT(mb_type)){
2551
                            s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2552
                        }else{
2553
                            s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2554
                        }
2555
                    }
2556
                }
2557
            }else{
2558
                for(i=0; i<16; i++){
2559
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2560
                        uint8_t * const ptr= dest_y + block_offset[i];
2561
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2562
                    }
2563
                }
2564
            }
2565
        }
2566

    
2567
        if((simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2568
            uint8_t *dest[2] = {dest_cb, dest_cr};
2569
            if(transform_bypass){
2570
                if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2571
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
2572
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
2573
                }else{
2574
                    idct_add = s->dsp.add_pixels4;
2575
                    for(i=16; i<16+8; i++){
2576
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2577
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2578
                    }
2579
                }
2580
            }else{
2581
                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]);
2582
                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]);
2583
                if(is_h264){
2584
                    idct_add = s->dsp.h264_idct_add;
2585
                    idct_dc_add = s->dsp.h264_idct_dc_add;
2586
                    for(i=16; i<16+8; i++){
2587
                        if(h->non_zero_count_cache[ scan8[i] ])
2588
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2589
                        else if(h->mb[i*16])
2590
                            idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2591
                    }
2592
                }else{
2593
                    for(i=16; i<16+8; i++){
2594
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2595
                            uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
2596
                            svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2597
                        }
2598
                    }
2599
                }
2600
            }
2601
        }
2602
    }
2603
    if(h->cbp || IS_INTRA(mb_type))
2604
        s->dsp.clear_blocks(h->mb);
2605

    
2606
    if(h->deblocking_filter) {
2607
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
2608
        fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
2609
        h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2610
        h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2611
        if (!simple && FRAME_MBAFF) {
2612
            filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2613
        } else {
2614
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2615
        }
2616
    }
2617
}
2618

    
2619
/**
2620
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2621
 */
2622
static void hl_decode_mb_simple(H264Context *h){
2623
    hl_decode_mb_internal(h, 1);
2624
}
2625

    
2626
/**
2627
 * Process a macroblock; this handles edge cases, such as interlacing.
2628
 */
2629
static void av_noinline hl_decode_mb_complex(H264Context *h){
2630
    hl_decode_mb_internal(h, 0);
2631
}
2632

    
2633
static void hl_decode_mb(H264Context *h){
2634
    MpegEncContext * const s = &h->s;
2635
    const int mb_xy= h->mb_xy;
2636
    const int mb_type= s->current_picture.mb_type[mb_xy];
2637
    int is_complex = ENABLE_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2638

    
2639
    if(ENABLE_H264_ENCODER && !s->decode)
2640
        return;
2641

    
2642
    if (is_complex)
2643
        hl_decode_mb_complex(h);
2644
    else hl_decode_mb_simple(h);
2645
}
2646

    
2647
static void pic_as_field(Picture *pic, const int parity){
2648
    int i;
2649
    for (i = 0; i < 4; ++i) {
2650
        if (parity == PICT_BOTTOM_FIELD)
2651
            pic->data[i] += pic->linesize[i];
2652
        pic->reference = parity;
2653
        pic->linesize[i] *= 2;
2654
    }
2655
    pic->poc= pic->field_poc[parity == PICT_BOTTOM_FIELD];
2656
}
2657

    
2658
static int split_field_copy(Picture *dest, Picture *src,
2659
                            int parity, int id_add){
2660
    int match = !!(src->reference & parity);
2661

    
2662
    if (match) {
2663
        *dest = *src;
2664
        if(parity != PICT_FRAME){
2665
            pic_as_field(dest, parity);
2666
            dest->pic_id *= 2;
2667
            dest->pic_id += id_add;
2668
        }
2669
    }
2670

    
2671
    return match;
2672
}
2673

    
2674
static int build_def_list(Picture *def, Picture **in, int len, int is_long, int sel){
2675
    int i[2]={0};
2676
    int index=0;
2677

    
2678
    while(i[0]<len || i[1]<len){
2679
        while(i[0]<len && !(in[ i[0] ] && (in[ i[0] ]->reference & sel)))
2680
            i[0]++;
2681
        while(i[1]<len && !(in[ i[1] ] && (in[ i[1] ]->reference & (sel^3))))
2682
            i[1]++;
2683
        if(i[0] < len){
2684
            in[ i[0] ]->pic_id= is_long ? i[0] : in[ i[0] ]->frame_num;
2685
            split_field_copy(&def[index++], in[ i[0]++ ], sel  , 1);
2686
        }
2687
        if(i[1] < len){
2688
            in[ i[1] ]->pic_id= is_long ? i[1] : in[ i[1] ]->frame_num;
2689
            split_field_copy(&def[index++], in[ i[1]++ ], sel^3, 0);
2690
        }
2691
    }
2692

    
2693
    return index;
2694
}
2695

    
2696
static int add_sorted(Picture **sorted, Picture **src, int len, int limit, int dir){
2697
    int i, best_poc;
2698
    int out_i= 0;
2699

    
2700
    for(;;){
2701
        best_poc= dir ? INT_MIN : INT_MAX;
2702

    
2703
        for(i=0; i<len; i++){
2704
            const int poc= src[i]->poc;
2705
            if(((poc > limit) ^ dir) && ((poc < best_poc) ^ dir)){
2706
                best_poc= poc;
2707
                sorted[out_i]= src[i];
2708
            }
2709
        }
2710
        if(best_poc == (dir ? INT_MIN : INT_MAX))
2711
            break;
2712
        limit= sorted[out_i++]->poc - dir;
2713
    }
2714
    return out_i;
2715
}
2716

    
2717
/**
2718
 * fills the default_ref_list.
2719
 */
2720
static int fill_default_ref_list(H264Context *h){
2721
    MpegEncContext * const s = &h->s;
2722
    int i, len;
2723

    
2724
    if(h->slice_type_nos==FF_B_TYPE){
2725
        Picture *sorted[32];
2726
        int cur_poc, list;
2727
        int lens[2];
2728

    
2729
        if(FIELD_PICTURE)
2730
            cur_poc= s->current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
2731
        else
2732
            cur_poc= s->current_picture_ptr->poc;
2733

    
2734
        for(list= 0; list<2; list++){
2735
            len= add_sorted(sorted    , h->short_ref, h->short_ref_count, cur_poc, 1^list);
2736
            len+=add_sorted(sorted+len, h->short_ref, h->short_ref_count, cur_poc, 0^list);
2737
            assert(len<=32);
2738
            len= build_def_list(h->default_ref_list[list]    , sorted     , len, 0, s->picture_structure);
2739
            len+=build_def_list(h->default_ref_list[list]+len, h->long_ref, 16 , 1, s->picture_structure);
2740
            assert(len<=32);
2741

    
2742
            if(len < h->ref_count[list])
2743
                memset(&h->default_ref_list[list][len], 0, sizeof(Picture)*(h->ref_count[list] - len));
2744
            lens[list]= len;
2745
        }
2746

    
2747
        if(lens[0] == lens[1] && lens[1] > 1){
2748
            for(i=0; h->default_ref_list[0][i].data[0] == h->default_ref_list[1][i].data[0] && i<lens[0]; i++);
2749
            if(i == lens[0])
2750
                FFSWAP(Picture, h->default_ref_list[1][0], h->default_ref_list[1][1]);
2751
        }
2752
    }else{
2753
        len = build_def_list(h->default_ref_list[0]    , h->short_ref, h->short_ref_count, 0, s->picture_structure);
2754
        len+= build_def_list(h->default_ref_list[0]+len, h-> long_ref, 16                , 1, s->picture_structure);
2755
        assert(len <= 32);
2756
        if(len < h->ref_count[0])
2757
            memset(&h->default_ref_list[0][len], 0, sizeof(Picture)*(h->ref_count[0] - len));
2758
    }
2759
#ifdef TRACE
2760
    for (i=0; i<h->ref_count[0]; i++) {
2761
        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]);
2762
    }
2763
    if(h->slice_type_nos==FF_B_TYPE){
2764
        for (i=0; i<h->ref_count[1]; i++) {
2765
            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]);
2766
        }
2767
    }
2768
#endif
2769
    return 0;
2770
}
2771

    
2772
static void print_short_term(H264Context *h);
2773
static void print_long_term(H264Context *h);
2774

    
2775
/**
2776
 * Extract structure information about the picture described by pic_num in
2777
 * the current decoding context (frame or field). Note that pic_num is
2778
 * picture number without wrapping (so, 0<=pic_num<max_pic_num).
2779
 * @param pic_num picture number for which to extract structure information
2780
 * @param structure one of PICT_XXX describing structure of picture
2781
 *                      with pic_num
2782
 * @return frame number (short term) or long term index of picture
2783
 *         described by pic_num
2784
 */
2785
static int pic_num_extract(H264Context *h, int pic_num, int *structure){
2786
    MpegEncContext * const s = &h->s;
2787

    
2788
    *structure = s->picture_structure;
2789
    if(FIELD_PICTURE){
2790
        if (!(pic_num & 1))
2791
            /* opposite field */
2792
            *structure ^= PICT_FRAME;
2793
        pic_num >>= 1;
2794
    }
2795

    
2796
    return pic_num;
2797
}
2798

    
2799
static int decode_ref_pic_list_reordering(H264Context *h){
2800
    MpegEncContext * const s = &h->s;
2801
    int list, index, pic_structure;
2802

    
2803
    print_short_term(h);
2804
    print_long_term(h);
2805

    
2806
    for(list=0; list<h->list_count; list++){
2807
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2808

    
2809
        if(get_bits1(&s->gb)){
2810
            int pred= h->curr_pic_num;
2811

    
2812
            for(index=0; ; index++){
2813
                unsigned int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2814
                unsigned int pic_id;
2815
                int i;
2816
                Picture *ref = NULL;
2817

    
2818
                if(reordering_of_pic_nums_idc==3)
2819
                    break;
2820

    
2821
                if(index >= h->ref_count[list]){
2822
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2823
                    return -1;
2824
                }
2825

    
2826
                if(reordering_of_pic_nums_idc<3){
2827
                    if(reordering_of_pic_nums_idc<2){
2828
                        const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2829
                        int frame_num;
2830

    
2831
                        if(abs_diff_pic_num > h->max_pic_num){
2832
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2833
                            return -1;
2834
                        }
2835

    
2836
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2837
                        else                                pred+= abs_diff_pic_num;
2838
                        pred &= h->max_pic_num - 1;
2839

    
2840
                        frame_num = pic_num_extract(h, pred, &pic_structure);
2841

    
2842
                        for(i= h->short_ref_count-1; i>=0; i--){
2843
                            ref = h->short_ref[i];
2844
                            assert(ref->reference);
2845
                            assert(!ref->long_ref);
2846
                            if(
2847
                                   ref->frame_num == frame_num &&
2848
                                   (ref->reference & pic_structure)
2849
                              )
2850
                                break;
2851
                        }
2852
                        if(i>=0)
2853
                            ref->pic_id= pred;
2854
                    }else{
2855
                        int long_idx;
2856
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2857

    
2858
                        long_idx= pic_num_extract(h, pic_id, &pic_structure);
2859

    
2860
                        if(long_idx>31){
2861
                            av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
2862
                            return -1;
2863
                        }
2864
                        ref = h->long_ref[long_idx];
2865
                        assert(!(ref && !ref->reference));
2866
                        if(ref && (ref->reference & pic_structure)){
2867
                            ref->pic_id= pic_id;
2868
                            assert(ref->long_ref);
2869
                            i=0;
2870
                        }else{
2871
                            i=-1;
2872
                        }
2873
                    }
2874

    
2875
                    if (i < 0) {
2876
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2877
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2878
                    } else {
2879
                        for(i=index; i+1<h->ref_count[list]; i++){
2880
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
2881
                                break;
2882
                        }
2883
                        for(; i > index; i--){
2884
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2885
                        }
2886
                        h->ref_list[list][index]= *ref;
2887
                        if (FIELD_PICTURE){
2888
                            pic_as_field(&h->ref_list[list][index], pic_structure);
2889
                        }
2890
                    }
2891
                }else{
2892
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2893
                    return -1;
2894
                }
2895
            }
2896
        }
2897
    }
2898
    for(list=0; list<h->list_count; list++){
2899
        for(index= 0; index < h->ref_count[list]; index++){
2900
            if(!h->ref_list[list][index].data[0]){
2901
                av_log(h->s.avctx, AV_LOG_ERROR, "Missing reference picture\n");
2902
                h->ref_list[list][index]= s->current_picture; //FIXME this is not a sensible solution
2903
            }
2904
        }
2905
    }
2906

    
2907
    return 0;
2908
}
2909

    
2910
static void fill_mbaff_ref_list(H264Context *h){
2911
    int list, i, j;
2912
    for(list=0; list<2; list++){ //FIXME try list_count
2913
        for(i=0; i<h->ref_count[list]; i++){
2914
            Picture *frame = &h->ref_list[list][i];
2915
            Picture *field = &h->ref_list[list][16+2*i];
2916
            field[0] = *frame;
2917
            for(j=0; j<3; j++)
2918
                field[0].linesize[j] <<= 1;
2919
            field[0].reference = PICT_TOP_FIELD;
2920
            field[0].poc= field[0].field_poc[0];
2921
            field[1] = field[0];
2922
            for(j=0; j<3; j++)
2923
                field[1].data[j] += frame->linesize[j];
2924
            field[1].reference = PICT_BOTTOM_FIELD;
2925
            field[1].poc= field[1].field_poc[1];
2926

    
2927
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
2928
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
2929
            for(j=0; j<2; j++){
2930
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
2931
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
2932
            }
2933
        }
2934
    }
2935
    for(j=0; j<h->ref_count[1]; j++){
2936
        for(i=0; i<h->ref_count[0]; i++)
2937
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
2938
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
2939
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
2940
    }
2941
}
2942

    
2943
static int pred_weight_table(H264Context *h){
2944
    MpegEncContext * const s = &h->s;
2945
    int list, i;
2946
    int luma_def, chroma_def;
2947

    
2948
    h->use_weight= 0;
2949
    h->use_weight_chroma= 0;
2950
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2951
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2952
    luma_def = 1<<h->luma_log2_weight_denom;
2953
    chroma_def = 1<<h->chroma_log2_weight_denom;
2954

    
2955
    for(list=0; list<2; list++){
2956
        for(i=0; i<h->ref_count[list]; i++){
2957
            int luma_weight_flag, chroma_weight_flag;
2958

    
2959
            luma_weight_flag= get_bits1(&s->gb);
2960
            if(luma_weight_flag){
2961
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
2962
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
2963
                if(   h->luma_weight[list][i] != luma_def
2964
                   || h->luma_offset[list][i] != 0)
2965
                    h->use_weight= 1;
2966
            }else{
2967
                h->luma_weight[list][i]= luma_def;
2968
                h->luma_offset[list][i]= 0;
2969
            }
2970

    
2971
            if(CHROMA){
2972
                chroma_weight_flag= get_bits1(&s->gb);
2973
                if(chroma_weight_flag){
2974
                    int j;
2975
                    for(j=0; j<2; j++){
2976
                        h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2977
                        h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2978
                        if(   h->chroma_weight[list][i][j] != chroma_def
2979
                        || h->chroma_offset[list][i][j] != 0)
2980
                            h->use_weight_chroma= 1;
2981
                    }
2982
                }else{
2983
                    int j;
2984
                    for(j=0; j<2; j++){
2985
                        h->chroma_weight[list][i][j]= chroma_def;
2986
                        h->chroma_offset[list][i][j]= 0;
2987
                    }
2988
                }
2989
            }
2990
        }
2991
        if(h->slice_type_nos != FF_B_TYPE) break;
2992
    }
2993
    h->use_weight= h->use_weight || h->use_weight_chroma;
2994
    return 0;
2995
}
2996

    
2997
static void implicit_weight_table(H264Context *h){
2998
    MpegEncContext * const s = &h->s;
2999
    int ref0, ref1;
3000
    int cur_poc = s->current_picture_ptr->poc;
3001

    
3002
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3003
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3004
        h->use_weight= 0;
3005
        h->use_weight_chroma= 0;
3006
        return;
3007
    }
3008

    
3009
    h->use_weight= 2;
3010
    h->use_weight_chroma= 2;
3011
    h->luma_log2_weight_denom= 5;
3012
    h->chroma_log2_weight_denom= 5;
3013

    
3014
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3015
        int poc0 = h->ref_list[0][ref0].poc;
3016
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3017
            int poc1 = h->ref_list[1][ref1].poc;
3018
            int td = av_clip(poc1 - poc0, -128, 127);
3019
            if(td){
3020
                int tb = av_clip(cur_poc - poc0, -128, 127);
3021
                int tx = (16384 + (FFABS(td) >> 1)) / td;
3022
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3023
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3024
                    h->implicit_weight[ref0][ref1] = 32;
3025
                else
3026
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3027
            }else
3028
                h->implicit_weight[ref0][ref1] = 32;
3029
        }
3030
    }
3031
}
3032

    
3033
/**
3034
 * Mark a picture as no longer needed for reference. The refmask
3035
 * argument allows unreferencing of individual fields or the whole frame.
3036
 * If the picture becomes entirely unreferenced, but is being held for
3037
 * display purposes, it is marked as such.
3038
 * @param refmask mask of fields to unreference; the mask is bitwise
3039
 *                anded with the reference marking of pic
3040
 * @return non-zero if pic becomes entirely unreferenced (except possibly
3041
 *         for display purposes) zero if one of the fields remains in
3042
 *         reference
3043
 */
3044
static inline int unreference_pic(H264Context *h, Picture *pic, int refmask){
3045
    int i;
3046
    if (pic->reference &= refmask) {
3047
        return 0;
3048
    } else {
3049
        for(i = 0; h->delayed_pic[i]; i++)
3050
            if(pic == h->delayed_pic[i]){
3051
                pic->reference=DELAYED_PIC_REF;
3052
                break;
3053
            }
3054
        return 1;
3055
    }
3056
}
3057

    
3058
/**
3059
 * instantaneous decoder refresh.
3060
 */
3061
static void idr(H264Context *h){
3062
    int i;
3063

    
3064
    for(i=0; i<16; i++){
3065
        remove_long(h, i, 0);
3066
    }
3067
    assert(h->long_ref_count==0);
3068

    
3069
    for(i=0; i<h->short_ref_count; i++){
3070
        unreference_pic(h, h->short_ref[i], 0);
3071
        h->short_ref[i]= NULL;
3072
    }
3073
    h->short_ref_count=0;
3074
    h->prev_frame_num= 0;
3075
    h->prev_frame_num_offset= 0;
3076
    h->prev_poc_msb=
3077
    h->prev_poc_lsb= 0;
3078
}
3079

    
3080
/* forget old pics after a seek */
3081
static void flush_dpb(AVCodecContext *avctx){
3082
    H264Context *h= avctx->priv_data;
3083
    int i;
3084
    for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
3085
        if(h->delayed_pic[i])
3086
            h->delayed_pic[i]->reference= 0;
3087
        h->delayed_pic[i]= NULL;
3088
    }
3089
    h->outputed_poc= INT_MIN;
3090
    idr(h);
3091
    if(h->s.current_picture_ptr)
3092
        h->s.current_picture_ptr->reference= 0;
3093
    h->s.first_field= 0;
3094
    ff_mpeg_flush(avctx);
3095
}
3096

    
3097
/**
3098
 * Find a Picture in the short term reference list by frame number.
3099
 * @param frame_num frame number to search for
3100
 * @param idx the index into h->short_ref where returned picture is found
3101
 *            undefined if no picture found.
3102
 * @return pointer to the found picture, or NULL if no pic with the provided
3103
 *                 frame number is found
3104
 */
3105
static Picture * find_short(H264Context *h, int frame_num, int *idx){
3106
    MpegEncContext * const s = &h->s;
3107
    int i;
3108

    
3109
    for(i=0; i<h->short_ref_count; i++){
3110
        Picture *pic= h->short_ref[i];
3111
        if(s->avctx->debug&FF_DEBUG_MMCO)
3112
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3113
        if(pic->frame_num == frame_num) {
3114
            *idx = i;
3115
            return pic;
3116
        }
3117
    }
3118
    return NULL;
3119
}
3120

    
3121
/**
3122
 * Remove a picture from the short term reference list by its index in
3123
 * that list.  This does no checking on the provided index; it is assumed
3124
 * to be valid. Other list entries are shifted down.
3125
 * @param i index into h->short_ref of picture to remove.
3126
 */
3127
static void remove_short_at_index(H264Context *h, int i){
3128
    assert(i >= 0 && i < h->short_ref_count);
3129
    h->short_ref[i]= NULL;
3130
    if (--h->short_ref_count)
3131
        memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i)*sizeof(Picture*));
3132
}
3133

    
3134
/**
3135
 *
3136
 * @return the removed picture or NULL if an error occurs
3137
 */
3138
static Picture * remove_short(H264Context *h, int frame_num, int ref_mask){
3139
    MpegEncContext * const s = &h->s;
3140
    Picture *pic;
3141
    int i;
3142

    
3143
    if(s->avctx->debug&FF_DEBUG_MMCO)
3144
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3145

    
3146
    pic = find_short(h, frame_num, &i);
3147
    if (pic){
3148
        if(unreference_pic(h, pic, ref_mask))
3149
        remove_short_at_index(h, i);
3150
    }
3151

    
3152
    return pic;
3153
}
3154

    
3155
/**
3156
 * Remove a picture from the long term reference list by its index in
3157
 * that list.
3158
 * @return the removed picture or NULL if an error occurs
3159
 */
3160
static Picture * remove_long(H264Context *h, int i, int ref_mask){
3161
    Picture *pic;
3162

    
3163
    pic= h->long_ref[i];
3164
    if (pic){
3165
        if(unreference_pic(h, pic, ref_mask)){
3166
            assert(h->long_ref[i]->long_ref == 1);
3167
            h->long_ref[i]->long_ref= 0;
3168
            h->long_ref[i]= NULL;
3169
            h->long_ref_count--;
3170
        }
3171
    }
3172

    
3173
    return pic;
3174
}
3175

    
3176
/**
3177
 * print short term list
3178
 */
3179
static void print_short_term(H264Context *h) {
3180
    uint32_t i;
3181
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3182
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3183
        for(i=0; i<h->short_ref_count; i++){
3184
            Picture *pic= h->short_ref[i];
3185
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3186
        }
3187
    }
3188
}
3189

    
3190
/**
3191
 * print long term list
3192
 */
3193
static void print_long_term(H264Context *h) {
3194
    uint32_t i;
3195
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3196
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3197
        for(i = 0; i < 16; i++){
3198
            Picture *pic= h->long_ref[i];
3199
            if (pic) {
3200
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3201
            }
3202
        }
3203
    }
3204
}
3205

    
3206
/**
3207
 * Executes the reference picture marking (memory management control operations).
3208
 */
3209
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3210
    MpegEncContext * const s = &h->s;
3211
    int i, j;
3212
    int current_ref_assigned=0;
3213
    Picture *pic;
3214

    
3215
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3216
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3217

    
3218
    for(i=0; i<mmco_count; i++){
3219
        int structure, frame_num;
3220
        if(s->avctx->debug&FF_DEBUG_MMCO)
3221
            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);
3222

    
3223
        if(   mmco[i].opcode == MMCO_SHORT2UNUSED
3224
           || mmco[i].opcode == MMCO_SHORT2LONG){
3225
            frame_num = pic_num_extract(h, mmco[i].short_pic_num, &structure);
3226
            pic = find_short(h, frame_num, &j);
3227
            if(!pic){
3228
                if(mmco[i].opcode != MMCO_SHORT2LONG || !h->long_ref[mmco[i].long_arg]
3229
                   || h->long_ref[mmco[i].long_arg]->frame_num != frame_num)
3230
                av_log(h->s.avctx, AV_LOG_ERROR, "mmco: unref short failure\n");
3231
                continue;
3232
            }
3233
        }
3234

    
3235
        switch(mmco[i].opcode){
3236
        case MMCO_SHORT2UNUSED:
3237
            if(s->avctx->debug&FF_DEBUG_MMCO)
3238
                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);
3239
            remove_short(h, frame_num, structure ^ PICT_FRAME);
3240
            break;
3241
        case MMCO_SHORT2LONG:
3242
                if (h->long_ref[mmco[i].long_arg] != pic)
3243
                    remove_long(h, mmco[i].long_arg, 0);
3244

    
3245
                remove_short_at_index(h, j);
3246
                h->long_ref[ mmco[i].long_arg ]= pic;
3247
                if (h->long_ref[ mmco[i].long_arg ]){
3248
                    h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3249
                    h->long_ref_count++;
3250
                }
3251
            break;
3252
        case MMCO_LONG2UNUSED:
3253
            j = pic_num_extract(h, mmco[i].long_arg, &structure);
3254
            pic = h->long_ref[j];
3255
            if (pic) {
3256
                remove_long(h, j, structure ^ PICT_FRAME);
3257
            } else if(s->avctx->debug&FF_DEBUG_MMCO)
3258
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref long failure\n");
3259
            break;
3260
        case MMCO_LONG:
3261
                    // Comment below left from previous code as it is an interresting note.
3262
                    /* First field in pair is in short term list or
3263
                     * at a different long term index.
3264
                     * This is not allowed; see 7.4.3.3, notes 2 and 3.
3265
                     * Report the problem and keep the pair where it is,
3266
                     * and mark this field valid.
3267
                     */
3268

    
3269
            if (h->long_ref[mmco[i].long_arg] != s->current_picture_ptr) {
3270
                remove_long(h, mmco[i].long_arg, 0);
3271

    
3272
                h->long_ref[ mmco[i].long_arg ]= s->current_picture_ptr;
3273
                h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3274
                h->long_ref_count++;
3275
            }
3276

    
3277
            s->current_picture_ptr->reference |= s->picture_structure;
3278
            current_ref_assigned=1;
3279
            break;
3280
        case MMCO_SET_MAX_LONG:
3281
            assert(mmco[i].long_arg <= 16);
3282
            // just remove the long term which index is greater than new max
3283
            for(j = mmco[i].long_arg; j<16; j++){
3284
                remove_long(h, j, 0);
3285
            }
3286
            break;
3287
        case MMCO_RESET:
3288
            while(h->short_ref_count){
3289
                remove_short(h, h->short_ref[0]->frame_num, 0);
3290
            }
3291
            for(j = 0; j < 16; j++) {
3292
                remove_long(h, j, 0);
3293
            }
3294
            s->current_picture_ptr->poc=
3295
            s->current_picture_ptr->field_poc[0]=
3296
            s->current_picture_ptr->field_poc[1]=
3297
            h->poc_lsb=
3298
            h->poc_msb=
3299
            h->frame_num=
3300
            s->current_picture_ptr->frame_num= 0;
3301
            break;
3302
        default: assert(0);
3303
        }
3304
    }
3305

    
3306
    if (!current_ref_assigned) {
3307
        /* Second field of complementary field pair; the first field of
3308
         * which is already referenced. If short referenced, it
3309
         * should be first entry in short_ref. If not, it must exist
3310
         * in long_ref; trying to put it on the short list here is an
3311
         * error in the encoded bit stream (ref: 7.4.3.3, NOTE 2 and 3).
3312
         */
3313
        if (h->short_ref_count && h->short_ref[0] == s->current_picture_ptr) {
3314
            /* Just mark the second field valid */
3315
            s->current_picture_ptr->reference = PICT_FRAME;
3316
        } else if (s->current_picture_ptr->long_ref) {
3317
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term reference "
3318
                                             "assignment for second field "
3319
                                             "in complementary field pair "
3320
                                             "(first field is long term)\n");
3321
        } else {
3322
            pic= remove_short(h, s->current_picture_ptr->frame_num, 0);
3323
            if(pic){
3324
                av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3325
            }
3326

    
3327
            if(h->short_ref_count)
3328
                memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3329

    
3330
            h->short_ref[0]= s->current_picture_ptr;
3331
            h->short_ref_count++;
3332
            s->current_picture_ptr->reference |= s->picture_structure;
3333
        }
3334
    }
3335

    
3336
    if (h->long_ref_count + h->short_ref_count > h->sps.ref_frame_count){
3337

    
3338
        /* We have too many reference frames, probably due to corrupted
3339
         * stream. Need to discard one frame. Prevents overrun of the
3340
         * short_ref and long_ref buffers.
3341
         */
3342
        av_log(h->s.avctx, AV_LOG_ERROR,
3343
               "number of reference frames exceeds max (probably "
3344
               "corrupt input), discarding one\n");
3345

    
3346
        if (h->long_ref_count && !h->short_ref_count) {
3347
            for (i = 0; i < 16; ++i)
3348
                if (h->long_ref[i])
3349
                    break;
3350

    
3351
            assert(i < 16);
3352
            remove_long(h, i, 0);
3353
        } else {
3354
            pic = h->short_ref[h->short_ref_count - 1];
3355
            remove_short(h, pic->frame_num, 0);
3356
        }
3357
    }
3358

    
3359
    print_short_term(h);
3360
    print_long_term(h);
3361
    return 0;
3362
}
3363

    
3364
static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
3365
    MpegEncContext * const s = &h->s;
3366
    int i;
3367

    
3368
    h->mmco_index= 0;
3369
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3370
        s->broken_link= get_bits1(gb) -1;
3371
        if(get_bits1(gb)){
3372
            h->mmco[0].opcode= MMCO_LONG;
3373
            h->mmco[0].long_arg= 0;
3374
            h->mmco_index= 1;
3375
        }
3376
    }else{
3377
        if(get_bits1(gb)){ // adaptive_ref_pic_marking_mode_flag
3378
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
3379
                MMCOOpcode opcode= get_ue_golomb(gb);
3380

    
3381
                h->mmco[i].opcode= opcode;
3382
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3383
                    h->mmco[i].short_pic_num= (h->curr_pic_num - get_ue_golomb(gb) - 1) & (h->max_pic_num - 1);
3384
/*                    if(h->mmco[i].short_pic_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_pic_num ] == NULL){
3385
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
3386
                        return -1;
3387
                    }*/
3388
                }
3389
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3390
                    unsigned int long_arg= get_ue_golomb(gb);
3391
                    if(long_arg >= 32 || (long_arg >= 16 && !(opcode == MMCO_LONG2UNUSED && FIELD_PICTURE))){
3392
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3393
                        return -1;
3394
                    }
3395
                    h->mmco[i].long_arg= long_arg;
3396
                }
3397

    
3398
                if(opcode > (unsigned)MMCO_LONG){
3399
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3400
                    return -1;
3401
                }
3402
                if(opcode == MMCO_END)
3403
                    break;
3404
            }
3405
            h->mmco_index= i;
3406
        }else{
3407
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3408

    
3409
            if(h->short_ref_count && h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count &&
3410
                    !(FIELD_PICTURE && !s->first_field && s->current_picture_ptr->reference)) {
3411
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3412
                h->mmco[0].short_pic_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3413
                h->mmco_index= 1;
3414
                if (FIELD_PICTURE) {
3415
                    h->mmco[0].short_pic_num *= 2;
3416
                    h->mmco[1].opcode= MMCO_SHORT2UNUSED;
3417
                    h->mmco[1].short_pic_num= h->mmco[0].short_pic_num + 1;
3418
                    h->mmco_index= 2;
3419
                }
3420
            }
3421
        }
3422
    }
3423

    
3424
    return 0;
3425
}
3426

    
3427
static int init_poc(H264Context *h){
3428
    MpegEncContext * const s = &h->s;
3429
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3430
    int field_poc[2];
3431
    Picture *cur = s->current_picture_ptr;
3432

    
3433
    h->frame_num_offset= h->prev_frame_num_offset;
3434
    if(h->frame_num < h->prev_frame_num)
3435
        h->frame_num_offset += max_frame_num;
3436

    
3437
    if(h->sps.poc_type==0){
3438
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3439

    
3440
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3441
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3442
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3443
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3444
        else
3445
            h->poc_msb = h->prev_poc_msb;
3446
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3447
        field_poc[0] =
3448
        field_poc[1] = h->poc_msb + h->poc_lsb;
3449
        if(s->picture_structure == PICT_FRAME)
3450
            field_poc[1] += h->delta_poc_bottom;
3451
    }else if(h->sps.poc_type==1){
3452
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3453
        int i;
3454

    
3455
        if(h->sps.poc_cycle_length != 0)
3456
            abs_frame_num = h->frame_num_offset + h->frame_num;
3457
        else
3458
            abs_frame_num = 0;
3459

    
3460
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
3461
            abs_frame_num--;
3462

    
3463
        expected_delta_per_poc_cycle = 0;
3464
        for(i=0; i < h->sps.poc_cycle_length; i++)
3465
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3466

    
3467
        if(abs_frame_num > 0){
3468
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3469
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3470

    
3471
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3472
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
3473
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3474
        } else
3475
            expectedpoc = 0;
3476

    
3477
        if(h->nal_ref_idc == 0)
3478
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3479

    
3480
        field_poc[0] = expectedpoc + h->delta_poc[0];
3481
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3482

    
3483
        if(s->picture_structure == PICT_FRAME)
3484
            field_poc[1] += h->delta_poc[1];
3485
    }else{
3486
        int poc= 2*(h->frame_num_offset + h->frame_num);
3487

    
3488
        if(!h->nal_ref_idc)
3489
            poc--;
3490

    
3491
        field_poc[0]= poc;
3492
        field_poc[1]= poc;
3493
    }
3494

    
3495
    if(s->picture_structure != PICT_BOTTOM_FIELD)
3496
        s->current_picture_ptr->field_poc[0]= field_poc[0];
3497
    if(s->picture_structure != PICT_TOP_FIELD)
3498
        s->current_picture_ptr->field_poc[1]= field_poc[1];
3499
    cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
3500

    
3501
    return 0;
3502
}
3503

    
3504

    
3505
/**
3506
 * initialize scan tables
3507
 */
3508
static void init_scan_tables(H264Context *h){
3509
    MpegEncContext * const s = &h->s;
3510
    int i;
3511
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
3512
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
3513
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
3514
    }else{
3515
        for(i=0; i<16; i++){
3516
#define T(x) (x>>2) | ((x<<2) & 0xF)
3517
            h->zigzag_scan[i] = T(zigzag_scan[i]);
3518
            h-> field_scan[i] = T( field_scan[i]);
3519
#undef T
3520
        }
3521
    }
3522
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
3523
        memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));
3524
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
3525
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
3526
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
3527
    }else{
3528
        for(i=0; i<64; i++){
3529
#define T(x) (x>>3) | ((x&7)<<3)
3530
            h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);
3531
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
3532
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
3533
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
3534
#undef T
3535
        }
3536
    }
3537
    if(h->sps.transform_bypass){ //FIXME same ugly
3538
        h->zigzag_scan_q0          = zigzag_scan;
3539
        h->zigzag_scan8x8_q0       = zigzag_scan8x8;
3540
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
3541
        h->field_scan_q0           = field_scan;
3542
        h->field_scan8x8_q0        = field_scan8x8;
3543
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
3544
    }else{
3545
        h->zigzag_scan_q0          = h->zigzag_scan;
3546
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
3547
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
3548
        h->field_scan_q0           = h->field_scan;
3549
        h->field_scan8x8_q0        = h->field_scan8x8;
3550
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
3551
    }
3552
}
3553

    
3554
/**
3555
 * Replicates H264 "master" context to thread contexts.
3556
 */
3557
static void clone_slice(H264Context *dst, H264Context *src)
3558
{
3559
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
3560
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
3561
    dst->s.current_picture      = src->s.current_picture;
3562
    dst->s.linesize             = src->s.linesize;
3563
    dst->s.uvlinesize           = src->s.uvlinesize;
3564
    dst->s.first_field          = src->s.first_field;
3565

    
3566
    dst->prev_poc_msb           = src->prev_poc_msb;
3567
    dst->prev_poc_lsb           = src->prev_poc_lsb;
3568
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
3569
    dst->prev_frame_num         = src->prev_frame_num;
3570
    dst->short_ref_count        = src->short_ref_count;
3571

    
3572
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
3573
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
3574
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
3575
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
3576

    
3577
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
3578
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
3579
}
3580

    
3581
/**
3582
 * decodes a slice header.
3583
 * This will also call MPV_common_init() and frame_start() as needed.
3584
 *
3585
 * @param h h264context
3586
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
3587
 *
3588
 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3589
 */
3590
static int decode_slice_header(H264Context *h, H264Context *h0){
3591
    MpegEncContext * const s = &h->s;
3592
    MpegEncContext * const s0 = &h0->s;
3593
    unsigned int first_mb_in_slice;
3594
    unsigned int pps_id;
3595
    int num_ref_idx_active_override_flag;
3596
    unsigned int slice_type, tmp, i, j;
3597
    int default_ref_list_done = 0;
3598
    int last_pic_structure;
3599

    
3600
    s->dropable= h->nal_ref_idc == 0;
3601

    
3602
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
3603
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
3604
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
3605
    }else{
3606
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
3607
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
3608
    }
3609

    
3610
    first_mb_in_slice= get_ue_golomb(&s->gb);
3611

    
3612
    if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
3613
        h0->current_slice = 0;
3614
        if (!s0->first_field)
3615
            s->current_picture_ptr= NULL;
3616
    }
3617

    
3618
    slice_type= get_ue_golomb(&s->gb);
3619
    if(slice_type > 9){
3620
        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);
3621
        return -1;
3622
    }
3623
    if(slice_type > 4){
3624
        slice_type -= 5;
3625
        h->slice_type_fixed=1;
3626
    }else
3627
        h->slice_type_fixed=0;
3628

    
3629
    slice_type= golomb_to_pict_type[ slice_type ];
3630
    if (slice_type == FF_I_TYPE
3631
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
3632
        default_ref_list_done = 1;
3633
    }
3634
    h->slice_type= slice_type;
3635
    h->slice_type_nos= slice_type & 3;
3636

    
3637
    s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
3638
    if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
3639
        av_log(h->s.avctx, AV_LOG_ERROR,
3640
               "B picture before any references, skipping\n");
3641
        return -1;
3642
    }
3643

    
3644
    pps_id= get_ue_golomb(&s->gb);
3645
    if(pps_id>=MAX_PPS_COUNT){
3646
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3647
        return -1;
3648
    }
3649
    if(!h0->pps_buffers[pps_id]) {
3650
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS referenced\n");
3651
        return -1;
3652
    }
3653
    h->pps= *h0->pps_buffers[pps_id];
3654

    
3655
    if(!h0->sps_buffers[h->pps.sps_id]) {
3656
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS referenced\n");
3657
        return -1;
3658
    }
3659
    h->sps = *h0->sps_buffers[h->pps.sps_id];
3660

    
3661
    if(h == h0 && h->dequant_coeff_pps != pps_id){
3662
        h->dequant_coeff_pps = pps_id;
3663
        init_dequant_tables(h);
3664
    }
3665

    
3666
    s->mb_width= h->sps.mb_width;
3667
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3668

    
3669
    h->b_stride=  s->mb_width*4;
3670
    h->b8_stride= s->mb_width*2;
3671

    
3672
    s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
3673
    if(h->sps.frame_mbs_only_flag)
3674
        s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
3675
    else
3676
        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
3677

    
3678
    if (s->context_initialized
3679
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3680
        if(h != h0)
3681
            return -1;   // width / height changed during parallelized decoding
3682
        free_tables(h);
3683
        flush_dpb(s->avctx);
3684
        MPV_common_end(s);
3685
    }
3686
    if (!s->context_initialized) {
3687
        if(h != h0)
3688
            return -1;  // we cant (re-)initialize context during parallel decoding
3689
        if (MPV_common_init(s) < 0)
3690
            return -1;
3691
        s->first_field = 0;
3692

    
3693
        init_scan_tables(h);
3694
        alloc_tables(h);
3695

    
3696
        for(i = 1; i < s->avctx->thread_count; i++) {
3697
            H264Context *c;
3698
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
3699
            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
3700
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
3701
            c->sps = h->sps;
3702
            c->pps = h->pps;
3703
            init_scan_tables(c);
3704
            clone_tables(c, h);
3705
        }
3706

    
3707
        for(i = 0; i < s->avctx->thread_count; i++)
3708
            if(context_init(h->thread_context[i]) < 0)
3709
                return -1;
3710

    
3711
        s->avctx->width = s->width;
3712
        s->avctx->height = s->height;
3713
        s->avctx->sample_aspect_ratio= h->sps.sar;
3714
        if(!s->avctx->sample_aspect_ratio.den)
3715
            s->avctx->sample_aspect_ratio.den = 1;
3716

    
3717
        if(h->sps.timing_info_present_flag){
3718
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale};
3719
            if(h->x264_build > 0 && h->x264_build < 44)
3720
                s->avctx->time_base.den *= 2;
3721
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3722
                      s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
3723
        }
3724
    }
3725

    
3726
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3727

    
3728
    h->mb_mbaff = 0;
3729
    h->mb_aff_frame = 0;
3730
    last_pic_structure = s0->picture_structure;
3731
    if(h->sps.frame_mbs_only_flag){
3732
        s->picture_structure= PICT_FRAME;
3733
    }else{
3734
        if(get_bits1(&s->gb)) { //field_pic_flag
3735
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3736
        } else {
3737
            s->picture_structure= PICT_FRAME;
3738
            h->mb_aff_frame = h->sps.mb_aff;
3739
        }
3740
    }
3741
    h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
3742

    
3743
    if(h0->current_slice == 0){
3744
        while(h->frame_num !=  h->prev_frame_num &&
3745
              h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
3746
            av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
3747
            frame_start(h);
3748
            h->prev_frame_num++;
3749
            h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
3750
            s->current_picture_ptr->frame_num= h->prev_frame_num;
3751
            execute_ref_pic_marking(h, NULL, 0);
3752
        }
3753

    
3754
        /* See if we have a decoded first field looking for a pair... */
3755
        if (s0->first_field) {
3756
            assert(s0->current_picture_ptr);
3757
            assert(s0->current_picture_ptr->data[0]);
3758
            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
3759

    
3760
            /* figure out if we have a complementary field pair */
3761
            if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
3762
                /*
3763
                 * Previous field is unmatched. Don't display it, but let it
3764
                 * remain for reference if marked as such.
3765
                 */
3766
                s0->current_picture_ptr = NULL;
3767
                s0->first_field = FIELD_PICTURE;
3768

    
3769
            } else {
3770
                if (h->nal_ref_idc &&
3771
                        s0->current_picture_ptr->reference &&
3772
                        s0->current_picture_ptr->frame_num != h->frame_num) {
3773
                    /*
3774
                     * This and previous field were reference, but had
3775
                     * different frame_nums. Consider this field first in
3776
                     * pair. Throw away previous field except for reference
3777
                     * purposes.
3778
                     */
3779
                    s0->first_field = 1;
3780
                    s0->current_picture_ptr = NULL;
3781

    
3782
                } else {
3783
                    /* Second field in complementary pair */
3784
                    s0->first_field = 0;
3785
                }
3786
            }
3787

    
3788
        } else {
3789
            /* Frame or first field in a potentially complementary pair */
3790
            assert(!s0->current_picture_ptr);
3791
            s0->first_field = FIELD_PICTURE;
3792
        }
3793

    
3794
        if((!FIELD_PICTURE || s0->first_field) && frame_start(h) < 0) {
3795
            s0->first_field = 0;
3796
            return -1;
3797
        }
3798
    }
3799
    if(h != h0)
3800
        clone_slice(h, h0);
3801

    
3802
    s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
3803

    
3804
    assert(s->mb_num == s->mb_width * s->mb_height);
3805
    if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
3806
       first_mb_in_slice                    >= s->mb_num){
3807
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3808
        return -1;
3809
    }
3810
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3811
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
3812
    if (s->picture_structure == PICT_BOTTOM_FIELD)
3813
        s->resync_mb_y = s->mb_y = s->mb_y + 1;
3814
    assert(s->mb_y < s->mb_height);
3815

    
3816
    if(s->picture_structure==PICT_FRAME){
3817
        h->curr_pic_num=   h->frame_num;
3818
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3819
    }else{
3820
        h->curr_pic_num= 2*h->frame_num + 1;
3821
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3822
    }
3823

    
3824
    if(h->nal_unit_type == NAL_IDR_SLICE){
3825
        get_ue_golomb(&s->gb); /* idr_pic_id */
3826
    }
3827

    
3828
    if(h->sps.poc_type==0){
3829
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3830

    
3831
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3832
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3833
        }
3834
    }
3835

    
3836
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3837
        h->delta_poc[0]= get_se_golomb(&s->gb);
3838

    
3839
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3840
            h->delta_poc[1]= get_se_golomb(&s->gb);
3841
    }
3842

    
3843
    init_poc(h);
3844

    
3845
    if(h->pps.redundant_pic_cnt_present){
3846
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3847
    }
3848

    
3849
    //set defaults, might be overridden a few lines later
3850
    h->ref_count[0]= h->pps.ref_count[0];
3851
    h->ref_count[1]= h->pps.ref_count[1];
3852

    
3853
    if(h->slice_type_nos != FF_I_TYPE){
3854
        if(h->slice_type_nos == FF_B_TYPE){
3855
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3856
        }
3857
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3858

    
3859
        if(num_ref_idx_active_override_flag){
3860
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3861
            if(h->slice_type_nos==FF_B_TYPE)
3862
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3863

    
3864
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
3865
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3866
                h->ref_count[0]= h->ref_count[1]= 1;
3867
                return -1;
3868
            }
3869
        }
3870
        if(h->slice_type_nos == FF_B_TYPE)
3871
            h->list_count= 2;
3872
        else
3873
            h->list_count= 1;
3874
    }else
3875
        h->list_count= 0;
3876

    
3877
    if(!default_ref_list_done){
3878
        fill_default_ref_list(h);
3879
    }
3880

    
3881
    if(h->slice_type_nos!=FF_I_TYPE && decode_ref_pic_list_reordering(h) < 0)
3882
        return -1;
3883

    
3884
    if(h->slice_type_nos!=FF_I_TYPE){
3885
        s->last_picture_ptr= &h->ref_list[0][0];
3886
        ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3887
    }
3888
    if(h->slice_type_nos==FF_B_TYPE){
3889
        s->next_picture_ptr= &h->ref_list[1][0];
3890
        ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3891
    }
3892

    
3893
    if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
3894
       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
3895
        pred_weight_table(h);
3896
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
3897
        implicit_weight_table(h);
3898
    else
3899
        h->use_weight = 0;
3900

    
3901
    if(h->nal_ref_idc)
3902
        decode_ref_pic_marking(h0, &s->gb);
3903

    
3904
    if(FRAME_MBAFF)
3905
        fill_mbaff_ref_list(h);
3906

    
3907
    if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
3908
        direct_dist_scale_factor(h);
3909
    direct_ref_list_init(h);
3910

    
3911
    if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
3912
        tmp = get_ue_golomb(&s->gb);
3913
        if(tmp > 2){
3914
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3915
            return -1;
3916
        }
3917
        h->cabac_init_idc= tmp;
3918
    }
3919

    
3920
    h->last_qscale_diff = 0;
3921
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3922
    if(tmp>51){
3923
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3924
        return -1;
3925
    }
3926
    s->qscale= tmp;
3927
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3928
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3929
    //FIXME qscale / qp ... stuff
3930
    if(h->slice_type == FF_SP_TYPE){
3931
        get_bits1(&s->gb); /* sp_for_switch_flag */
3932
    }
3933
    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
3934
        get_se_golomb(&s->gb); /* slice_qs_delta */
3935
    }
3936

    
3937
    h->deblocking_filter = 1;
3938
    h->slice_alpha_c0_offset = 0;
3939
    h->slice_beta_offset = 0;
3940
    if( h->pps.deblocking_filter_parameters_present ) {
3941
        tmp= get_ue_golomb(&s->gb);
3942
        if(tmp > 2){
3943
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3944
            return -1;
3945
        }
3946
        h->deblocking_filter= tmp;
3947
        if(h->deblocking_filter < 2)
3948
            h->deblocking_filter^= 1; // 1<->0
3949

    
3950
        if( h->deblocking_filter ) {
3951
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3952
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3953
        }
3954
    }
3955

    
3956
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
3957
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
3958
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
3959
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3960
        h->deblocking_filter= 0;
3961

    
3962
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
3963
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3964
            /* Cheat slightly for speed:
3965
               Do not bother to deblock across slices. */
3966
            h->deblocking_filter = 2;
3967
        } else {
3968
            h0->max_contexts = 1;
3969
            if(!h0->single_decode_warning) {
3970
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3971
                h0->single_decode_warning = 1;
3972
            }
3973
            if(h != h0)
3974
                return 1; // deblocking switched inside frame
3975
        }
3976
    }
3977

    
3978
#if 0 //FMO
3979
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3980
        slice_group_change_cycle= get_bits(&s->gb, ?);
3981
#endif
3982

    
3983
    h0->last_slice_type = slice_type;
3984
    h->slice_num = ++h0->current_slice;
3985
    if(h->slice_num >= MAX_SLICES){
3986
        av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
3987
    }
3988

    
3989
    for(j=0; j<2; j++){
3990
        int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
3991
        ref2frm[0]=
3992
        ref2frm[1]= -1;
3993
        for(i=0; i<16; i++)
3994
            ref2frm[i+2]= 4*h->ref_list[j][i].frame_num
3995
                          +(h->ref_list[j][i].reference&3);
3996
        ref2frm[18+0]=
3997
        ref2frm[18+1]= -1;
3998
        for(i=16; i<48; i++)
3999
            ref2frm[i+4]= 4*h->ref_list[j][i].frame_num
4000
                          +(h->ref_list[j][i].reference&3);
4001
    }
4002

    
4003
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
4004
    h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
4005

    
4006
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4007
        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",
4008
               h->slice_num,
4009
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
4010
               first_mb_in_slice,
4011
               av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
4012
               pps_id, h->frame_num,
4013
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4014
               h->ref_count[0], h->ref_count[1],
4015
               s->qscale,
4016
               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4017
               h->use_weight,
4018
               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
4019
               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
4020
               );
4021
    }
4022

    
4023
    return 0;
4024
}
4025

    
4026
/**
4027
 *
4028
 */
4029
static inline int get_level_prefix(GetBitContext *gb){
4030
    unsigned int buf;
4031
    int log;
4032

    
4033
    OPEN_READER(re, gb);
4034
    UPDATE_CACHE(re, gb);
4035
    buf=GET_CACHE(re, gb);
4036

    
4037
    log= 32 - av_log2(buf);
4038
#ifdef TRACE
4039
    print_bin(buf>>(32-log), log);
4040
    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__);
4041
#endif
4042

    
4043
    LAST_SKIP_BITS(re, gb, log);
4044
    CLOSE_READER(re, gb);
4045

    
4046
    return log-1;
4047
}
4048

    
4049
static inline int get_dct8x8_allowed(H264Context *h){
4050
    if(h->sps.direct_8x8_inference_flag)
4051
        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
4052
    else
4053
        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
4054
}
4055

    
4056
/**
4057
 * decodes a residual block.
4058
 * @param n block index
4059
 * @param scantable scantable
4060
 * @param max_coeff number of coefficients in the block
4061
 * @return <0 if an error occurred
4062
 */
4063
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
4064
    MpegEncContext * const s = &h->s;
4065
    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};
4066
    int level[16];
4067
    int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;
4068

    
4069
    //FIXME put trailing_onex into the context
4070

    
4071
    if(n == CHROMA_DC_BLOCK_INDEX){
4072
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4073
        total_coeff= coeff_token>>2;
4074
    }else{
4075
        if(n == LUMA_DC_BLOCK_INDEX){
4076
            total_coeff= pred_non_zero_count(h, 0);
4077
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4078
            total_coeff= coeff_token>>2;
4079
        }else{
4080
            total_coeff= pred_non_zero_count(h, n);
4081
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4082
            total_coeff= coeff_token>>2;
4083
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4084
        }
4085
    }
4086

    
4087
    //FIXME set last_non_zero?
4088

    
4089
    if(total_coeff==0)
4090
        return 0;
4091
    if(total_coeff > (unsigned)max_coeff) {
4092
        av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
4093
        return -1;
4094
    }
4095

    
4096
    trailing_ones= coeff_token&3;
4097
    tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
4098
    assert(total_coeff<=16);
4099

    
4100
    i = show_bits(gb, 3);
4101
    skip_bits(gb, trailing_ones);
4102
    level[0] = 1-((i&4)>>1);
4103
    level[1] = 1-((i&2)   );
4104
    level[2] = 1-((i&1)<<1);
4105

    
4106
    if(trailing_ones<total_coeff) {
4107
        int level_code, mask;
4108
        int suffix_length = total_coeff > 10 && trailing_ones < 3;
4109
        int prefix= get_level_prefix(gb);
4110

    
4111
        //first coefficient has suffix_length equal to 0 or 1
4112
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4113
            if(suffix_length)
4114
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4115
            else
4116
                level_code= (prefix<<suffix_length); //part
4117
        }else if(prefix==14){
4118
            if(suffix_length)
4119
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4120
            else
4121
                level_code= prefix + get_bits(gb, 4); //part
4122
        }else{
4123
            level_code= (15<<suffix_length) + get_bits(gb, prefix-3); //part
4124
            if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
4125
            if(prefix>=16)
4126
                level_code += (1<<(prefix-3))-4096;
4127
        }
4128

    
4129
        if(trailing_ones < 3) level_code += 2;
4130

    
4131
        suffix_length = 1;
4132
        if(level_code > 5)
4133
            suffix_length++;
4134
        mask= -(level_code&1);
4135
        level[trailing_ones]= (((2+level_code)>>1) ^ mask) - mask;
4136

    
4137
        //remaining coefficients have suffix_length > 0
4138
        for(i=trailing_ones+1;i<total_coeff;i++) {
4139
            static const int suffix_limit[7] = {0,5,11,23,47,95,INT_MAX };
4140
            prefix = get_level_prefix(gb);
4141
            if(prefix<15){
4142
                level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
4143
            }else{
4144
                level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
4145
                if(prefix>=16)
4146
                    level_code += (1<<(prefix-3))-4096;
4147
            }
4148
            mask= -(level_code&1);
4149
            level[i]= (((2+level_code)>>1) ^ mask) - mask;
4150
            if(level_code > suffix_limit[suffix_length])
4151
                suffix_length++;
4152
        }
4153
    }
4154

    
4155
    if(total_coeff == max_coeff)
4156
        zeros_left=0;
4157
    else{
4158
        if(n == CHROMA_DC_BLOCK_INDEX)
4159
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4160
        else
4161
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4162
    }
4163

    
4164
    coeff_num = zeros_left + total_coeff - 1;
4165
    j = scantable[coeff_num];
4166
    if(n > 24){
4167
        block[j] = level[0];
4168
        for(i=1;i<total_coeff;i++) {
4169
            if(zeros_left <= 0)
4170
                run_before = 0;
4171
            else if(zeros_left < 7){
4172
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4173
            }else{
4174
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4175
            }
4176
            zeros_left -= run_before;
4177
            coeff_num -= 1 + run_before;
4178
            j= scantable[ coeff_num ];
4179

    
4180
            block[j]= level[i];
4181
        }
4182
    }else{
4183
        block[j] = (level[0] * qmul[j] + 32)>>6;
4184
        for(i=1;i<total_coeff;i++) {
4185
            if(zeros_left <= 0)
4186
                run_before = 0;
4187
            else if(zeros_left < 7){
4188
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4189
            }else{
4190
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4191
            }
4192
            zeros_left -= run_before;
4193
            coeff_num -= 1 + run_before;
4194
            j= scantable[ coeff_num ];
4195

    
4196
            block[j]= (level[i] * qmul[j] + 32)>>6;
4197
        }
4198
    }
4199

    
4200
    if(zeros_left<0){
4201
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4202
        return -1;
4203
    }
4204

    
4205
    return 0;
4206
}
4207

    
4208
static void predict_field_decoding_flag(H264Context *h){
4209
    MpegEncContext * const s = &h->s;
4210
    const int mb_xy= h->mb_xy;
4211
    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
4212
                ? s->current_picture.mb_type[mb_xy-1]
4213
                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
4214
                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
4215
                : 0;
4216
    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4217
}
4218

    
4219
/**
4220
 * decodes a P_SKIP or B_SKIP macroblock
4221
 */
4222
static void decode_mb_skip(H264Context *h){
4223
    MpegEncContext * const s = &h->s;
4224
    const int mb_xy= h->mb_xy;
4225
    int mb_type=0;
4226

    
4227
    memset(h->non_zero_count[mb_xy], 0, 16);
4228
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4229

    
4230
    if(MB_FIELD)
4231
        mb_type|= MB_TYPE_INTERLACED;
4232

    
4233
    if( h->slice_type_nos == FF_B_TYPE )
4234
    {
4235
        // just for fill_caches. pred_direct_motion will set the real mb_type
4236
        mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4237

    
4238
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4239
        pred_direct_motion(h, &mb_type);
4240
        mb_type|= MB_TYPE_SKIP;
4241
    }
4242
    else
4243
    {
4244
        int mx, my;
4245
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4246

    
4247
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4248
        pred_pskip_motion(h, &mx, &my);
4249
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4250
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4251
    }
4252

    
4253
    write_back_motion(h, mb_type);
4254
    s->current_picture.mb_type[mb_xy]= mb_type;
4255
    s->current_picture.qscale_table[mb_xy]= s->qscale;
4256
    h->slice_table[ mb_xy ]= h->slice_num;
4257
    h->prev_mb_skipped= 1;
4258
}
4259

    
4260
/**
4261
 * decodes a macroblock
4262
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4263
 */
4264
static int decode_mb_cavlc(H264Context *h){
4265
    MpegEncContext * const s = &h->s;
4266
    int mb_xy;
4267
    int partition_count;
4268
    unsigned int mb_type, cbp;
4269
    int dct8x8_allowed= h->pps.transform_8x8_mode;
4270

    
4271
    mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4272

    
4273
    tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4274
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4275
                down the code */
4276
    if(h->slice_type_nos != FF_I_TYPE){
4277
        if(s->mb_skip_run==-1)
4278
            s->mb_skip_run= get_ue_golomb(&s->gb);
4279

    
4280
        if (s->mb_skip_run--) {
4281
            if(FRAME_MBAFF && (s->mb_y&1) == 0){
4282
                if(s->mb_skip_run==0)
4283
                    h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4284
                else
4285
                    predict_field_decoding_flag(h);
4286
            }
4287
            decode_mb_skip(h);
4288
            return 0;
4289
        }
4290
    }
4291
    if(FRAME_MBAFF){
4292
        if( (s->mb_y&1) == 0 )
4293
            h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4294
    }
4295

    
4296
    h->prev_mb_skipped= 0;
4297

    
4298
    mb_type= get_ue_golomb(&s->gb);
4299
    if(h->slice_type_nos == FF_B_TYPE){
4300
        if(mb_type < 23){
4301
            partition_count= b_mb_type_info[mb_type].partition_count;
4302
            mb_type=         b_mb_type_info[mb_type].type;
4303
        }else{
4304
            mb_type -= 23;
4305
            goto decode_intra_mb;
4306
        }
4307
    }else if(h->slice_type_nos == FF_P_TYPE){
4308
        if(mb_type < 5){
4309
            partition_count= p_mb_type_info[mb_type].partition_count;
4310
            mb_type=         p_mb_type_info[mb_type].type;
4311
        }else{
4312
            mb_type -= 5;
4313
            goto decode_intra_mb;
4314
        }
4315
    }else{
4316
       assert(h->slice_type_nos == FF_I_TYPE);
4317
        if(h->slice_type == FF_SI_TYPE && mb_type)
4318
            mb_type--;
4319
decode_intra_mb:
4320
        if(mb_type > 25){
4321
            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);
4322
            return -1;
4323
        }
4324
        partition_count=0;
4325
        cbp= i_mb_type_info[mb_type].cbp;
4326
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4327
        mb_type= i_mb_type_info[mb_type].type;
4328
    }
4329

    
4330
    if(MB_FIELD)
4331
        mb_type |= MB_TYPE_INTERLACED;
4332

    
4333
    h->slice_table[ mb_xy ]= h->slice_num;
4334

    
4335
    if(IS_INTRA_PCM(mb_type)){
4336
        unsigned int x;
4337

    
4338
        // We assume these blocks are very rare so we do not optimize it.
4339
        align_get_bits(&s->gb);
4340

    
4341
        // The pixels are stored in the same order as levels in h->mb array.
4342
        for(x=0; x < (CHROMA ? 384 : 256); x++){
4343
            ((uint8_t*)h->mb)[x]= get_bits(&s->gb, 8);
4344
        }
4345

    
4346
        // In deblocking, the quantizer is 0
4347
        s->current_picture.qscale_table[mb_xy]= 0;
4348
        // All coeffs are present
4349
        memset(h->non_zero_count[mb_xy], 16, 16);
4350

    
4351
        s->current_picture.mb_type[mb_xy]= mb_type;
4352
        return 0;
4353
    }
4354

    
4355
    if(MB_MBAFF){
4356
        h->ref_count[0] <<= 1;
4357
        h->ref_count[1] <<= 1;
4358
    }
4359

    
4360
    fill_caches(h, mb_type, 0);
4361

    
4362
    //mb_pred
4363
    if(IS_INTRA(mb_type)){
4364
        int pred_mode;
4365
//            init_top_left_availability(h);
4366
        if(IS_INTRA4x4(mb_type)){
4367
            int i;
4368
            int di = 1;
4369
            if(dct8x8_allowed && get_bits1(&s->gb)){
4370
                mb_type |= MB_TYPE_8x8DCT;
4371
                di = 4;
4372
            }
4373

    
4374
//                fill_intra4x4_pred_table(h);
4375
            for(i=0; i<16; i+=di){
4376
                int mode= pred_intra_mode(h, i);
4377

    
4378
                if(!get_bits1(&s->gb)){
4379
                    const int rem_mode= get_bits(&s->gb, 3);
4380
                    mode = rem_mode + (rem_mode >= mode);
4381
                }
4382

    
4383
                if(di==4)
4384
                    fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
4385
                else
4386
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4387
            }
4388
            write_back_intra_pred_mode(h);
4389
            if( check_intra4x4_pred_mode(h) < 0)
4390
                return -1;
4391
        }else{
4392
            h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4393
            if(h->intra16x16_pred_mode < 0)
4394
                return -1;
4395
        }
4396
        if(CHROMA){
4397
            pred_mode= check_intra_pred_mode(h, get_ue_golomb(&s->gb));
4398
            if(pred_mode < 0)
4399
                return -1;
4400
            h->chroma_pred_mode= pred_mode;
4401
        }
4402
    }else if(partition_count==4){
4403
        int i, j, sub_partition_count[4], list, ref[2][4];
4404

    
4405
        if(h->slice_type_nos == FF_B_TYPE){
4406
            for(i=0; i<4; i++){
4407
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4408
                if(h->sub_mb_type[i] >=13){
4409
                    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);
4410
                    return -1;
4411
                }
4412
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4413
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4414
            }
4415
            if(   IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4416
               || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
4417
                pred_direct_motion(h, &mb_type);
4418
                h->ref_cache[0][scan8[4]] =
4419
                h->ref_cache[1][scan8[4]] =
4420
                h->ref_cache[0][scan8[12]] =
4421
                h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
4422
            }
4423
        }else{
4424
            assert(h->slice_type_nos == FF_P_TYPE); //FIXME SP correct ?
4425
            for(i=0; i<4; i++){
4426
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4427
                if(h->sub_mb_type[i] >=4){
4428
                    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);
4429
                    return -1;
4430
                }
4431
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4432
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4433
            }
4434
        }
4435

    
4436
        for(list=0; list<h->list_count; list++){
4437
            int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4438
            for(i=0; i<4; i++){
4439
                if(IS_DIRECT(h->sub_mb_type[i])) continue;
4440
                if(IS_DIR(h->sub_mb_type[i], 0, list)){
4441
                    unsigned int tmp = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4442
                    if(tmp>=ref_count){
4443
                        av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", tmp);
4444
                        return -1;
4445
                    }
4446
                    ref[list][i]= tmp;
4447
                }else{
4448
                 //FIXME
4449
                    ref[list][i] = -1;
4450
                }
4451
            }
4452
        }
4453

    
4454
        if(dct8x8_allowed)
4455
            dct8x8_allowed = get_dct8x8_allowed(h);
4456

    
4457
        for(list=0; list<h->list_count; list++){
4458
            for(i=0; i<4; i++){
4459
                if(IS_DIRECT(h->sub_mb_type[i])) {
4460
                    h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ];
4461
                    continue;
4462
                }
4463
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
<