Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ ebb61f83

History | View | Annotate | Download (314 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 libavcodec/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 "mathops.h"
36
#include "rectangle.h"
37
#include "vdpau_internal.h"
38

    
39
#include "cabac.h"
40
#if ARCH_X86
41
#include "x86/h264_i386.h"
42
#endif
43

    
44
//#undef NDEBUG
45
#include <assert.h>
46

    
47
/**
48
 * Value of Picture.reference when Picture is not a reference picture, but
49
 * is held for delayed output.
50
 */
51
#define DELAYED_PIC_REF 4
52

    
53
static VLC coeff_token_vlc[4];
54
static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
55
static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
56

    
57
static VLC chroma_dc_coeff_token_vlc;
58
static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
59
static const int chroma_dc_coeff_token_vlc_table_size = 256;
60

    
61
static VLC total_zeros_vlc[15];
62
static VLC_TYPE total_zeros_vlc_tables[15][512][2];
63
static const int total_zeros_vlc_tables_size = 512;
64

    
65
static VLC chroma_dc_total_zeros_vlc[3];
66
static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
67
static const int chroma_dc_total_zeros_vlc_tables_size = 8;
68

    
69
static VLC run_vlc[6];
70
static VLC_TYPE run_vlc_tables[6][8][2];
71
static const int run_vlc_tables_size = 8;
72

    
73
static VLC run7_vlc;
74
static VLC_TYPE run7_vlc_table[96][2];
75
static const int run7_vlc_table_size = 96;
76

    
77
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
78
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
79
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);
80
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);
81
static Picture * remove_long(H264Context *h, int i, int ref_mask);
82

    
83
static av_always_inline uint32_t pack16to32(int a, int b){
84
#ifdef WORDS_BIGENDIAN
85
   return (b&0xFFFF) + (a<<16);
86
#else
87
   return (a&0xFFFF) + (b<<16);
88
#endif
89
}
90

    
91
static const uint8_t rem6[52]={
92
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,
93
};
94

    
95
static const uint8_t div6[52]={
96
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,
97
};
98

    
99
static const uint8_t left_block_options[4][8]={
100
    {0,1,2,3,7,10,8,11},
101
    {2,2,3,3,8,11,8,11},
102
    {0,0,1,1,7,10,7,10},
103
    {0,2,0,2,7,10,7,10}
104
};
105

    
106
#define LEVEL_TAB_BITS 8
107
static int8_t cavlc_level_tab[7][1<<LEVEL_TAB_BITS][2];
108

    
109
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
110
    MpegEncContext * const s = &h->s;
111
    const int mb_xy= h->mb_xy;
112
    int topleft_xy, top_xy, topright_xy, left_xy[2];
113
    int topleft_type, top_type, topright_type, left_type[2];
114
    const uint8_t * left_block;
115
    int topleft_partition= -1;
116
    int i;
117

    
118
    top_xy     = mb_xy  - (s->mb_stride << FIELD_PICTURE);
119

    
120
    //FIXME deblocking could skip the intra and nnz parts.
121
    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
122
        return;
123

    
124
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
125
     * stuff, I can't imagine that these complex rules are worth it. */
126

    
127
    topleft_xy = top_xy - 1;
128
    topright_xy= top_xy + 1;
129
    left_xy[1] = left_xy[0] = mb_xy-1;
130
    left_block = left_block_options[0];
131
    if(FRAME_MBAFF){
132
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
133
        const int top_pair_xy      = pair_xy     - s->mb_stride;
134
        const int topleft_pair_xy  = top_pair_xy - 1;
135
        const int topright_pair_xy = top_pair_xy + 1;
136
        const int topleft_mb_field_flag  = IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
137
        const int top_mb_field_flag      = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
138
        const int topright_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
139
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
140
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
141
        const int bottom = (s->mb_y & 1);
142
        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);
143

    
144
        if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
145
            top_xy -= s->mb_stride;
146
        }
147
        if (curr_mb_field_flag && (bottom || topleft_mb_field_flag)){
148
            topleft_xy -= s->mb_stride;
149
        } else if(bottom && !curr_mb_field_flag && left_mb_field_flag) {
150
            topleft_xy += s->mb_stride;
151
            // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
152
            topleft_partition = 0;
153
        }
154
        if (curr_mb_field_flag && (bottom || topright_mb_field_flag)){
155
            topright_xy -= s->mb_stride;
156
        }
157
        if (left_mb_field_flag != curr_mb_field_flag) {
158
            left_xy[1] = left_xy[0] = pair_xy - 1;
159
            if (curr_mb_field_flag) {
160
                left_xy[1] += s->mb_stride;
161
                left_block = left_block_options[3];
162
            } else {
163
                left_block= left_block_options[2 - bottom];
164
            }
165
        }
166
    }
167

    
168
    h->top_mb_xy = top_xy;
169
    h->left_mb_xy[0] = left_xy[0];
170
    h->left_mb_xy[1] = left_xy[1];
171
    if(for_deblock){
172
        topleft_type = 0;
173
        topright_type = 0;
174
        top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
175
        left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
176
        left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
177

    
178
        if(MB_MBAFF && !IS_INTRA(mb_type)){
179
            int list;
180
            for(list=0; list<h->list_count; list++){
181
                //These values where changed for ease of performing MC, we need to change them back
182
                //FIXME maybe we can make MC and loop filter use the same values or prevent
183
                //the MC code from changing ref_cache and rather use a temporary array.
184
                if(USES_LIST(mb_type,list)){
185
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
186
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
187
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
188
                    ref += h->b8_stride;
189
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
190
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
191
                }
192
            }
193
        }
194
    }else{
195
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
196
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
197
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
198
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
199
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
200

    
201
    if(IS_INTRA(mb_type)){
202
        int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
203
        h->topleft_samples_available=
204
        h->top_samples_available=
205
        h->left_samples_available= 0xFFFF;
206
        h->topright_samples_available= 0xEEEA;
207

    
208
        if(!(top_type & type_mask)){
209
            h->topleft_samples_available= 0xB3FF;
210
            h->top_samples_available= 0x33FF;
211
            h->topright_samples_available= 0x26EA;
212
        }
213
        if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
214
            if(IS_INTERLACED(mb_type)){
215
                if(!(left_type[0] & type_mask)){
216
                    h->topleft_samples_available&= 0xDFFF;
217
                    h->left_samples_available&= 0x5FFF;
218
                }
219
                if(!(left_type[1] & type_mask)){
220
                    h->topleft_samples_available&= 0xFF5F;
221
                    h->left_samples_available&= 0xFF5F;
222
                }
223
            }else{
224
                int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
225
                                ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
226
                assert(left_xy[0] == left_xy[1]);
227
                if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
228
                    h->topleft_samples_available&= 0xDF5F;
229
                    h->left_samples_available&= 0x5F5F;
230
                }
231
            }
232
        }else{
233
            if(!(left_type[0] & type_mask)){
234
                h->topleft_samples_available&= 0xDF5F;
235
                h->left_samples_available&= 0x5F5F;
236
            }
237
        }
238

    
239
        if(!(topleft_type & type_mask))
240
            h->topleft_samples_available&= 0x7FFF;
241

    
242
        if(!(topright_type & type_mask))
243
            h->topright_samples_available&= 0xFBFF;
244

    
245
        if(IS_INTRA4x4(mb_type)){
246
            if(IS_INTRA4x4(top_type)){
247
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
248
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
249
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
250
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
251
            }else{
252
                int pred;
253
                if(!(top_type & type_mask))
254
                    pred= -1;
255
                else{
256
                    pred= 2;
257
                }
258
                h->intra4x4_pred_mode_cache[4+8*0]=
259
                h->intra4x4_pred_mode_cache[5+8*0]=
260
                h->intra4x4_pred_mode_cache[6+8*0]=
261
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
262
            }
263
            for(i=0; i<2; i++){
264
                if(IS_INTRA4x4(left_type[i])){
265
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
266
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
267
                }else{
268
                    int pred;
269
                    if(!(left_type[i] & type_mask))
270
                        pred= -1;
271
                    else{
272
                        pred= 2;
273
                    }
274
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
275
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
276
                }
277
            }
278
        }
279
    }
280
    }
281

    
282

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

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

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

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

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

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

    
316
    }
317

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

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

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

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

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

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

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

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

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

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

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

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

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

    
493
                    if(IS_DIRECT(left_type[0]))
494
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
495
                    else if(IS_8X8(left_type[0]))
496
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
497
                    else
498
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
499

    
500
                    if(IS_DIRECT(left_type[1]))
501
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
502
                    else if(IS_8X8(left_type[1]))
503
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
504
                    else
505
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
506
                }
507
            }
508

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

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

    
548
static inline void write_back_intra_pred_mode(H264Context *h){
549
    const int mb_xy= h->mb_xy;
550

    
551
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
552
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
553
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
554
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
555
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
556
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
557
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
558
}
559

    
560
/**
561
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
562
 */
563
static inline int check_intra4x4_pred_mode(H264Context *h){
564
    MpegEncContext * const s = &h->s;
565
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
566
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
567
    int i;
568

    
569
    if(!(h->top_samples_available&0x8000)){
570
        for(i=0; i<4; i++){
571
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
572
            if(status<0){
573
                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);
574
                return -1;
575
            } else if(status){
576
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
577
            }
578
        }
579
    }
580

    
581
    if((h->left_samples_available&0x8888)!=0x8888){
582
        static const int mask[4]={0x8000,0x2000,0x80,0x20};
583
        for(i=0; i<4; i++){
584
            if(!(h->left_samples_available&mask[i])){
585
                int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
586
                if(status<0){
587
                    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);
588
                    return -1;
589
                } else if(status){
590
                    h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
591
                }
592
            }
593
        }
594
    }
595

    
596
    return 0;
597
} //FIXME cleanup like next
598

    
599
/**
600
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
601
 */
602
static inline int check_intra_pred_mode(H264Context *h, int mode){
603
    MpegEncContext * const s = &h->s;
604
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
605
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
606

    
607
    if(mode > 6U) {
608
        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);
609
        return -1;
610
    }
611

    
612
    if(!(h->top_samples_available&0x8000)){
613
        mode= top[ mode ];
614
        if(mode<0){
615
            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);
616
            return -1;
617
        }
618
    }
619

    
620
    if((h->left_samples_available&0x8080) != 0x8080){
621
        mode= left[ mode ];
622
        if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
623
            mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
624
        }
625
        if(mode<0){
626
            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);
627
            return -1;
628
        }
629
    }
630

    
631
    return mode;
632
}
633

    
634
/**
635
 * gets the predicted intra4x4 prediction mode.
636
 */
637
static inline int pred_intra_mode(H264Context *h, int n){
638
    const int index8= scan8[n];
639
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
640
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
641
    const int min= FFMIN(left, top);
642

    
643
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
644

    
645
    if(min<0) return DC_PRED;
646
    else      return min;
647
}
648

    
649
static inline void write_back_non_zero_count(H264Context *h){
650
    const int mb_xy= h->mb_xy;
651

    
652
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
653
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
654
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
655
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
656
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
657
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
658
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
659

    
660
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
661
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
662
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
663

    
664
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
665
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
666
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
667
}
668

    
669
/**
670
 * gets the predicted number of non-zero coefficients.
671
 * @param n block index
672
 */
673
static inline int pred_non_zero_count(H264Context *h, int n){
674
    const int index8= scan8[n];
675
    const int left= h->non_zero_count_cache[index8 - 1];
676
    const int top = h->non_zero_count_cache[index8 - 8];
677
    int i= left + top;
678

    
679
    if(i<64) i= (i+1)>>1;
680

    
681
    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
682

    
683
    return i&31;
684
}
685

    
686
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
687
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
688
    MpegEncContext *s = &h->s;
689

    
690
    /* there is no consistent mapping of mvs to neighboring locations that will
691
     * make mbaff happy, so we can't move all this logic to fill_caches */
692
    if(FRAME_MBAFF){
693
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
694
        const int16_t *mv;
695
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
696
        *C = h->mv_cache[list][scan8[0]-2];
697

    
698
        if(!MB_FIELD
699
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
700
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
701
            if(IS_INTERLACED(mb_types[topright_xy])){
702
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
703
                const int x4 = X4, y4 = Y4;\
704
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
705
                if(!USES_LIST(mb_type,list))\
706
                    return LIST_NOT_USED;\
707
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
708
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
709
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
710
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
711

    
712
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
713
            }
714
        }
715
        if(topright_ref == PART_NOT_AVAILABLE
716
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
717
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
718
            if(!MB_FIELD
719
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
720
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
721
            }
722
            if(MB_FIELD
723
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
724
               && i >= scan8[0]+8){
725
                // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
726
                SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
727
            }
728
        }
729
#undef SET_DIAG_MV
730
    }
731

    
732
    if(topright_ref != PART_NOT_AVAILABLE){
733
        *C= h->mv_cache[list][ i - 8 + part_width ];
734
        return topright_ref;
735
    }else{
736
        tprintf(s->avctx, "topright MV not available\n");
737

    
738
        *C= h->mv_cache[list][ i - 8 - 1 ];
739
        return h->ref_cache[list][ i - 8 - 1 ];
740
    }
741
}
742

    
743
/**
744
 * gets the predicted MV.
745
 * @param n the block index
746
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
747
 * @param mx the x component of the predicted motion vector
748
 * @param my the y component of the predicted motion vector
749
 */
750
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
751
    const int index8= scan8[n];
752
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
753
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
754
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
755
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
756
    const int16_t * C;
757
    int diagonal_ref, match_count;
758

    
759
    assert(part_width==1 || part_width==2 || part_width==4);
760

    
761
/* mv_cache
762
  B . . A T T T T
763
  U . . L . . , .
764
  U . . L . . . .
765
  U . . L . . , .
766
  . . . L . . . .
767
*/
768

    
769
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
770
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
771
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
772
    if(match_count > 1){ //most common
773
        *mx= mid_pred(A[0], B[0], C[0]);
774
        *my= mid_pred(A[1], B[1], C[1]);
775
    }else if(match_count==1){
776
        if(left_ref==ref){
777
            *mx= A[0];
778
            *my= A[1];
779
        }else if(top_ref==ref){
780
            *mx= B[0];
781
            *my= B[1];
782
        }else{
783
            *mx= C[0];
784
            *my= C[1];
785
        }
786
    }else{
787
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
788
            *mx= A[0];
789
            *my= A[1];
790
        }else{
791
            *mx= mid_pred(A[0], B[0], C[0]);
792
            *my= mid_pred(A[1], B[1], C[1]);
793
        }
794
    }
795

    
796
    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);
797
}
798

    
799
/**
800
 * gets the directionally predicted 16x8 MV.
801
 * @param n the block index
802
 * @param mx the x component of the predicted motion vector
803
 * @param my the y component of the predicted motion vector
804
 */
805
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
806
    if(n==0){
807
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
808
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
809

    
810
        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);
811

    
812
        if(top_ref == ref){
813
            *mx= B[0];
814
            *my= B[1];
815
            return;
816
        }
817
    }else{
818
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
819
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
820

    
821
        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);
822

    
823
        if(left_ref == ref){
824
            *mx= A[0];
825
            *my= A[1];
826
            return;
827
        }
828
    }
829

    
830
    //RARE
831
    pred_motion(h, n, 4, list, ref, mx, my);
832
}
833

    
834
/**
835
 * gets the directionally predicted 8x16 MV.
836
 * @param n the block index
837
 * @param mx the x component of the predicted motion vector
838
 * @param my the y component of the predicted motion vector
839
 */
840
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
841
    if(n==0){
842
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
843
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
844

    
845
        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);
846

    
847
        if(left_ref == ref){
848
            *mx= A[0];
849
            *my= A[1];
850
            return;
851
        }
852
    }else{
853
        const int16_t * C;
854
        int diagonal_ref;
855

    
856
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
857

    
858
        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);
859

    
860
        if(diagonal_ref == ref){
861
            *mx= C[0];
862
            *my= C[1];
863
            return;
864
        }
865
    }
866

    
867
    //RARE
868
    pred_motion(h, n, 2, list, ref, mx, my);
869
}
870

    
871
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
872
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
873
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
874

    
875
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
876

    
877
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
878
       || !( top_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ])
879
       || !(left_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ])){
880

    
881
        *mx = *my = 0;
882
        return;
883
    }
884

    
885
    pred_motion(h, 0, 4, 0, 0, mx, my);
886

    
887
    return;
888
}
889

    
890
static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
891
    int poc0 = h->ref_list[0][i].poc;
892
    int td = av_clip(poc1 - poc0, -128, 127);
893
    if(td == 0 || h->ref_list[0][i].long_ref){
894
        return 256;
895
    }else{
896
        int tb = av_clip(poc - poc0, -128, 127);
897
        int tx = (16384 + (FFABS(td) >> 1)) / td;
898
        return av_clip((tb*tx + 32) >> 6, -1024, 1023);
899
    }
900
}
901

    
902
static inline void direct_dist_scale_factor(H264Context * const h){
903
    MpegEncContext * const s = &h->s;
904
    const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
905
    const int poc1 = h->ref_list[1][0].poc;
906
    int i, field;
907
    for(field=0; field<2; field++){
908
        const int poc  = h->s.current_picture_ptr->field_poc[field];
909
        const int poc1 = h->ref_list[1][0].field_poc[field];
910
        for(i=0; i < 2*h->ref_count[0]; i++)
911
            h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
912
    }
913

    
914
    for(i=0; i<h->ref_count[0]; i++){
915
        h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
916
    }
917
}
918

    
919
static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
920
    MpegEncContext * const s = &h->s;
921
    Picture * const ref1 = &h->ref_list[1][0];
922
    int j, old_ref, rfield;
923
    int start= mbafi ? 16                      : 0;
924
    int end  = mbafi ? 16+2*h->ref_count[list] : h->ref_count[list];
925
    int interl= mbafi || s->picture_structure != PICT_FRAME;
926

    
927
    /* bogus; fills in for missing frames */
928
    memset(map[list], 0, sizeof(map[list]));
929

    
930
    for(rfield=0; rfield<2; rfield++){
931
        for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
932
            int poc = ref1->ref_poc[colfield][list][old_ref];
933

    
934
            if     (!interl)
935
                poc |= 3;
936
            else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
937
                poc= (poc&~3) + rfield + 1;
938

    
939
            for(j=start; j<end; j++){
940
                if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
941
                    int cur_ref= mbafi ? (j-16)^field : j;
942
                    map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
943
                    if(rfield == field)
944
                        map[list][old_ref] = cur_ref;
945
                    break;
946
                }
947
            }
948
        }
949
    }
950
}
951

    
952
static inline void direct_ref_list_init(H264Context * const h){
953
    MpegEncContext * const s = &h->s;
954
    Picture * const ref1 = &h->ref_list[1][0];
955
    Picture * const cur = s->current_picture_ptr;
956
    int list, j, field;
957
    int sidx= (s->picture_structure&1)^1;
958
    int ref1sidx= (ref1->reference&1)^1;
959

    
960
    for(list=0; list<2; list++){
961
        cur->ref_count[sidx][list] = h->ref_count[list];
962
        for(j=0; j<h->ref_count[list]; j++)
963
            cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
964
    }
965

    
966
    if(s->picture_structure == PICT_FRAME){
967
        memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
968
        memcpy(cur->ref_poc  [1], cur->ref_poc  [0], sizeof(cur->ref_poc  [0]));
969
    }
970

    
971
    cur->mbaff= FRAME_MBAFF;
972

    
973
    if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
974
        return;
975

    
976
    for(list=0; list<2; list++){
977
        fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
978
        for(field=0; field<2; field++)
979
            fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
980
    }
981
}
982

    
983
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
984
    MpegEncContext * const s = &h->s;
985
    int b8_stride = h->b8_stride;
986
    int b4_stride = h->b_stride;
987
    int mb_xy = h->mb_xy;
988
    int mb_type_col[2];
989
    const int16_t (*l1mv0)[2], (*l1mv1)[2];
990
    const int8_t *l1ref0, *l1ref1;
991
    const int is_b8x8 = IS_8X8(*mb_type);
992
    unsigned int sub_mb_type;
993
    int i8, i4;
994

    
995
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
996

    
997
    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
998
        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
999
            int cur_poc = s->current_picture_ptr->poc;
1000
            int *col_poc = h->ref_list[1]->field_poc;
1001
            int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
1002
            mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
1003
            b8_stride = 0;
1004
        }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity
1005
            int fieldoff= 2*(h->ref_list[1][0].reference)-3;
1006
            mb_xy += s->mb_stride*fieldoff;
1007
        }
1008
        goto single_col;
1009
    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
1010
        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
1011
            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
1012
            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
1013
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
1014
            b8_stride *= 3;
1015
            b4_stride *= 6;
1016
            //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag
1017
            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
1018
                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
1019
                && !is_b8x8){
1020
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1021
                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
1022
            }else{
1023
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1024
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1025
            }
1026
        }else{                                           //     AFR/FR    -> AFR/FR
1027
single_col:
1028
            mb_type_col[0] =
1029
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
1030
            if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
1031
                /* FIXME save sub mb types from previous frames (or derive from MVs)
1032
                * so we know exactly what block size to use */
1033
                sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1034
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1035
            }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
1036
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1037
                *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1038
            }else{
1039
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1040
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1041
            }
1042
        }
1043
    }
1044

    
1045
    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
1046
    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
1047
    l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
1048
    l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
1049
    if(!b8_stride){
1050
        if(s->mb_y&1){
1051
            l1ref0 += h->b8_stride;
1052
            l1ref1 += h->b8_stride;
1053
            l1mv0  +=  2*b4_stride;
1054
            l1mv1  +=  2*b4_stride;
1055
        }
1056
    }
1057

    
1058
    if(h->direct_spatial_mv_pred){
1059
        int ref[2];
1060
        int mv[2][2];
1061
        int list;
1062

    
1063
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1064

    
1065
        /* ref = min(neighbors) */
1066
        for(list=0; list<2; list++){
1067
            int refa = h->ref_cache[list][scan8[0] - 1];
1068
            int refb = h->ref_cache[list][scan8[0] - 8];
1069
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1070
            if(refc == PART_NOT_AVAILABLE)
1071
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1072
            ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
1073
            if(ref[list] < 0)
1074
                ref[list] = -1;
1075
        }
1076

    
1077
        if(ref[0] < 0 && ref[1] < 0){
1078
            ref[0] = ref[1] = 0;
1079
            mv[0][0] = mv[0][1] =
1080
            mv[1][0] = mv[1][1] = 0;
1081
        }else{
1082
            for(list=0; list<2; list++){
1083
                if(ref[list] >= 0)
1084
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1085
                else
1086
                    mv[list][0] = mv[list][1] = 0;
1087
            }
1088
        }
1089

    
1090
        if(ref[1] < 0){
1091
            if(!is_b8x8)
1092
                *mb_type &= ~MB_TYPE_L1;
1093
            sub_mb_type &= ~MB_TYPE_L1;
1094
        }else if(ref[0] < 0){
1095
            if(!is_b8x8)
1096
                *mb_type &= ~MB_TYPE_L0;
1097
            sub_mb_type &= ~MB_TYPE_L0;
1098
        }
1099

    
1100
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1101
            for(i8=0; i8<4; i8++){
1102
                int x8 = i8&1;
1103
                int y8 = i8>>1;
1104
                int xy8 = x8+y8*b8_stride;
1105
                int xy4 = 3*x8+y8*b4_stride;
1106
                int a=0, b=0;
1107

    
1108
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1109
                    continue;
1110
                h->sub_mb_type[i8] = sub_mb_type;
1111

    
1112
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1113
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1114
                if(!IS_INTRA(mb_type_col[y8])
1115
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1116
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1117
                    if(ref[0] > 0)
1118
                        a= pack16to32(mv[0][0],mv[0][1]);
1119
                    if(ref[1] > 0)
1120
                        b= pack16to32(mv[1][0],mv[1][1]);
1121
                }else{
1122
                    a= pack16to32(mv[0][0],mv[0][1]);
1123
                    b= pack16to32(mv[1][0],mv[1][1]);
1124
                }
1125
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1126
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1127
            }
1128
        }else if(IS_16X16(*mb_type)){
1129
            int a=0, b=0;
1130

    
1131
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1132
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1133
            if(!IS_INTRA(mb_type_col[0])
1134
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1135
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1136
                       && (h->x264_build>33 || !h->x264_build)))){
1137
                if(ref[0] > 0)
1138
                    a= pack16to32(mv[0][0],mv[0][1]);
1139
                if(ref[1] > 0)
1140
                    b= pack16to32(mv[1][0],mv[1][1]);
1141
            }else{
1142
                a= pack16to32(mv[0][0],mv[0][1]);
1143
                b= pack16to32(mv[1][0],mv[1][1]);
1144
            }
1145
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1146
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1147
        }else{
1148
            for(i8=0; i8<4; i8++){
1149
                const int x8 = i8&1;
1150
                const int y8 = i8>>1;
1151

    
1152
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1153
                    continue;
1154
                h->sub_mb_type[i8] = sub_mb_type;
1155

    
1156
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1157
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1158
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1159
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1160

    
1161
                /* col_zero_flag */
1162
                if(!IS_INTRA(mb_type_col[0]) && (   l1ref0[x8 + y8*b8_stride] == 0
1163
                                              || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
1164
                                                  && (h->x264_build>33 || !h->x264_build)))){
1165
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
1166
                    if(IS_SUB_8X8(sub_mb_type)){
1167
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1168
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1169
                            if(ref[0] == 0)
1170
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1171
                            if(ref[1] == 0)
1172
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1173
                        }
1174
                    }else
1175
                    for(i4=0; i4<4; i4++){
1176
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1177
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1178
                            if(ref[0] == 0)
1179
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1180
                            if(ref[1] == 0)
1181
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1182
                        }
1183
                    }
1184
                }
1185
            }
1186
        }
1187
    }else{ /* direct temporal mv pred */
1188
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1189
        const int *dist_scale_factor = h->dist_scale_factor;
1190
        int ref_offset= 0;
1191

    
1192
        if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
1193
            map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
1194
            map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
1195
            dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];
1196
        }
1197
        if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
1198
            ref_offset += 16;
1199

    
1200
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1201
            /* FIXME assumes direct_8x8_inference == 1 */
1202
            int y_shift  = 2*!IS_INTERLACED(*mb_type);
1203

    
1204
            for(i8=0; i8<4; i8++){
1205
                const int x8 = i8&1;
1206
                const int y8 = i8>>1;
1207
                int ref0, scale;
1208
                const int16_t (*l1mv)[2]= l1mv0;
1209

    
1210
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1211
                    continue;
1212
                h->sub_mb_type[i8] = sub_mb_type;
1213

    
1214
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1215
                if(IS_INTRA(mb_type_col[y8])){
1216
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1217
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1218
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1219
                    continue;
1220
                }
1221

    
1222
                ref0 = l1ref0[x8 + y8*b8_stride];
1223
                if(ref0 >= 0)
1224
                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
1225
                else{
1226
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1227
                    l1mv= l1mv1;
1228
                }
1229
                scale = dist_scale_factor[ref0];
1230
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1231

    
1232
                {
1233
                    const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
1234
                    int my_col = (mv_col[1]<<y_shift)/2;
1235
                    int mx = (scale * mv_col[0] + 128) >> 8;
1236
                    int my = (scale * my_col + 128) >> 8;
1237
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1238
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1239
                }
1240
            }
1241
            return;
1242
        }
1243

    
1244
        /* one-to-one mv scaling */
1245

    
1246
        if(IS_16X16(*mb_type)){
1247
            int ref, mv0, mv1;
1248

    
1249
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1250
            if(IS_INTRA(mb_type_col[0])){
1251
                ref=mv0=mv1=0;
1252
            }else{
1253
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
1254
                                                : map_col_to_list0[1][l1ref1[0] + ref_offset];
1255
                const int scale = dist_scale_factor[ref0];
1256
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1257
                int mv_l0[2];
1258
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1259
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1260
                ref= ref0;
1261
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
1262
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1263
            }
1264
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1265
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1266
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1267
        }else{
1268
            for(i8=0; i8<4; i8++){
1269
                const int x8 = i8&1;
1270
                const int y8 = i8>>1;
1271
                int ref0, scale;
1272
                const int16_t (*l1mv)[2]= l1mv0;
1273

    
1274
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1275
                    continue;
1276
                h->sub_mb_type[i8] = sub_mb_type;
1277
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1278
                if(IS_INTRA(mb_type_col[0])){
1279
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1280
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1281
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1282
                    continue;
1283
                }
1284

    
1285
                ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset;
1286
                if(ref0 >= 0)
1287
                    ref0 = map_col_to_list0[0][ref0];
1288
                else{
1289
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1290
                    l1mv= l1mv1;
1291
                }
1292
                scale = dist_scale_factor[ref0];
1293

    
1294
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1295
                if(IS_SUB_8X8(sub_mb_type)){
1296
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1297
                    int mx = (scale * mv_col[0] + 128) >> 8;
1298
                    int my = (scale * mv_col[1] + 128) >> 8;
1299
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1300
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1301
                }else
1302
                for(i4=0; i4<4; i4++){
1303
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1304
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1305
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1306
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1307
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1308
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1309
                }
1310
            }
1311
        }
1312
    }
1313
}
1314

    
1315
static inline void write_back_motion(H264Context *h, int mb_type){
1316
    MpegEncContext * const s = &h->s;
1317
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1318
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1319
    int list;
1320

    
1321
    if(!USES_LIST(mb_type, 0))
1322
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1323

    
1324
    for(list=0; list<h->list_count; list++){
1325
        int y;
1326
        if(!USES_LIST(mb_type, list))
1327
            continue;
1328

    
1329
        for(y=0; y<4; y++){
1330
            *(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];
1331
            *(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];
1332
        }
1333
        if( h->pps.cabac ) {
1334
            if(IS_SKIP(mb_type))
1335
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1336
            else
1337
            for(y=0; y<4; y++){
1338
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1339
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1340
            }
1341
        }
1342

    
1343
        {
1344
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1345
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1346
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1347
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1348
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1349
        }
1350
    }
1351

    
1352
    if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
1353
        if(IS_8X8(mb_type)){
1354
            uint8_t *direct_table = &h->direct_table[b8_xy];
1355
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1356
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1357
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1358
        }
1359
    }
1360
}
1361

    
1362
/**
1363
 * Decodes a network abstraction layer unit.
1364
 * @param consumed is the number of bytes used as input
1365
 * @param length is the length of the array
1366
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1367
 * @returns decoded bytes, might be src+1 if no escapes
1368
 */
1369
static const uint8_t *decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1370
    int i, si, di;
1371
    uint8_t *dst;
1372
    int bufidx;
1373

    
1374
//    src[0]&0x80;                //forbidden bit
1375
    h->nal_ref_idc= src[0]>>5;
1376
    h->nal_unit_type= src[0]&0x1F;
1377

    
1378
    src++; length--;
1379
#if 0
1380
    for(i=0; i<length; i++)
1381
        printf("%2X ", src[i]);
1382
#endif
1383

    
1384
#if HAVE_FAST_UNALIGNED
1385
# if HAVE_FAST_64BIT
1386
#   define RS 7
1387
    for(i=0; i+1<length; i+=9){
1388
        if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
1389
# else
1390
#   define RS 3
1391
    for(i=0; i+1<length; i+=5){
1392
        if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
1393
# endif
1394
            continue;
1395
        if(i>0 && !src[i]) i--;
1396
        while(src[i]) i++;
1397
#else
1398
#   define RS 0
1399
    for(i=0; i+1<length; i+=2){
1400
        if(src[i]) continue;
1401
        if(i>0 && src[i-1]==0) i--;
1402
#endif
1403
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1404
            if(src[i+2]!=3){
1405
                /* startcode, so we must be past the end */
1406
                length=i;
1407
            }
1408
            break;
1409
        }
1410
        i-= RS;
1411
    }
1412

    
1413
    if(i>=length-1){ //no escaped 0
1414
        *dst_length= length;
1415
        *consumed= length+1; //+1 for the header
1416
        return src;
1417
    }
1418

    
1419
    bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1420
    h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
1421
    dst= h->rbsp_buffer[bufidx];
1422

    
1423
    if (dst == NULL){
1424
        return NULL;
1425
    }
1426

    
1427
//printf("decoding esc\n");
1428
    memcpy(dst, src, i);
1429
    si=di=i;
1430
    while(si+2<length){
1431
        //remove escapes (very rare 1:2^22)
1432
        if(src[si+2]>3){
1433
            dst[di++]= src[si++];
1434
            dst[di++]= src[si++];
1435
        }else if(src[si]==0 && src[si+1]==0){
1436
            if(src[si+2]==3){ //escape
1437
                dst[di++]= 0;
1438
                dst[di++]= 0;
1439
                si+=3;
1440
                continue;
1441
            }else //next start code
1442
                goto nsc;
1443
        }
1444

    
1445
        dst[di++]= src[si++];
1446
    }
1447
    while(si<length)
1448
        dst[di++]= src[si++];
1449
nsc:
1450

    
1451
    memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
1452

    
1453
    *dst_length= di;
1454
    *consumed= si + 1;//+1 for the header
1455
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1456
    return dst;
1457
}
1458

    
1459
/**
1460
 * identifies the exact end of the bitstream
1461
 * @return the length of the trailing, or 0 if damaged
1462
 */
1463
static int decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1464
    int v= *src;
1465
    int r;
1466

    
1467
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1468

    
1469
    for(r=1; r<9; r++){
1470
        if(v&1) return r;
1471
        v>>=1;
1472
    }
1473
    return 0;
1474
}
1475

    
1476
/**
1477
 * IDCT transforms the 16 dc values and dequantizes them.
1478
 * @param qp quantization parameter
1479
 */
1480
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1481
#define stride 16
1482
    int i;
1483
    int temp[16]; //FIXME check if this is a good idea
1484
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1485
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1486

    
1487
//memset(block, 64, 2*256);
1488
//return;
1489
    for(i=0; i<4; i++){
1490
        const int offset= y_offset[i];
1491
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1492
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1493
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1494
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1495

    
1496
        temp[4*i+0]= z0+z3;
1497
        temp[4*i+1]= z1+z2;
1498
        temp[4*i+2]= z1-z2;
1499
        temp[4*i+3]= z0-z3;
1500
    }
1501

    
1502
    for(i=0; i<4; i++){
1503
        const int offset= x_offset[i];
1504
        const int z0= temp[4*0+i] + temp[4*2+i];
1505
        const int z1= temp[4*0+i] - temp[4*2+i];
1506
        const int z2= temp[4*1+i] - temp[4*3+i];
1507
        const int z3= temp[4*1+i] + temp[4*3+i];
1508

    
1509
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
1510
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1511
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1512
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1513
    }
1514
}
1515

    
1516
#if 0
1517
/**
1518
 * DCT transforms the 16 dc values.
1519
 * @param qp quantization parameter ??? FIXME
1520
 */
1521
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1522
//    const int qmul= dequant_coeff[qp][0];
1523
    int i;
1524
    int temp[16]; //FIXME check if this is a good idea
1525
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1526
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1527

1528
    for(i=0; i<4; i++){
1529
        const int offset= y_offset[i];
1530
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1531
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1532
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1533
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1534

1535
        temp[4*i+0]= z0+z3;
1536
        temp[4*i+1]= z1+z2;
1537
        temp[4*i+2]= z1-z2;
1538
        temp[4*i+3]= z0-z3;
1539
    }
1540

1541
    for(i=0; i<4; i++){
1542
        const int offset= x_offset[i];
1543
        const int z0= temp[4*0+i] + temp[4*2+i];
1544
        const int z1= temp[4*0+i] - temp[4*2+i];
1545
        const int z2= temp[4*1+i] - temp[4*3+i];
1546
        const int z3= temp[4*1+i] + temp[4*3+i];
1547

1548
        block[stride*0 +offset]= (z0 + z3)>>1;
1549
        block[stride*2 +offset]= (z1 + z2)>>1;
1550
        block[stride*8 +offset]= (z1 - z2)>>1;
1551
        block[stride*10+offset]= (z0 - z3)>>1;
1552
    }
1553
}
1554
#endif
1555

    
1556
#undef xStride
1557
#undef stride
1558

    
1559
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1560
    const int stride= 16*2;
1561
    const int xStride= 16;
1562
    int a,b,c,d,e;
1563

    
1564
    a= block[stride*0 + xStride*0];
1565
    b= block[stride*0 + xStride*1];
1566
    c= block[stride*1 + xStride*0];
1567
    d= block[stride*1 + xStride*1];
1568

    
1569
    e= a-b;
1570
    a= a+b;
1571
    b= c-d;
1572
    c= c+d;
1573

    
1574
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1575
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1576
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1577
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1578
}
1579

    
1580
#if 0
1581
static void chroma_dc_dct_c(DCTELEM *block){
1582
    const int stride= 16*2;
1583
    const int xStride= 16;
1584
    int a,b,c,d,e;
1585

1586
    a= block[stride*0 + xStride*0];
1587
    b= block[stride*0 + xStride*1];
1588
    c= block[stride*1 + xStride*0];
1589
    d= block[stride*1 + xStride*1];
1590

1591
    e= a-b;
1592
    a= a+b;
1593
    b= c-d;
1594
    c= c+d;
1595

1596
    block[stride*0 + xStride*0]= (a+c);
1597
    block[stride*0 + xStride*1]= (e+b);
1598
    block[stride*1 + xStride*0]= (a-c);
1599
    block[stride*1 + xStride*1]= (e-b);
1600
}
1601
#endif
1602

    
1603
/**
1604
 * gets the chroma qp.
1605
 */
1606
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1607
    return h->pps.chroma_qp_table[t][qscale];
1608
}
1609

    
1610
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1611
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1612
                           int src_x_offset, int src_y_offset,
1613
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1614
    MpegEncContext * const s = &h->s;
1615
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1616
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1617
    const int luma_xy= (mx&3) + ((my&3)<<2);
1618
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1619
    uint8_t * src_cb, * src_cr;
1620
    int extra_width= h->emu_edge_width;
1621
    int extra_height= h->emu_edge_height;
1622
    int emu=0;
1623
    const int full_mx= mx>>2;
1624
    const int full_my= my>>2;
1625
    const int pic_width  = 16*s->mb_width;
1626
    const int pic_height = 16*s->mb_height >> MB_FIELD;
1627

    
1628
    if(mx&7) extra_width -= 3;
1629
    if(my&7) extra_height -= 3;
1630

    
1631
    if(   full_mx < 0-extra_width
1632
       || full_my < 0-extra_height
1633
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
1634
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
1635
        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);
1636
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1637
        emu=1;
1638
    }
1639

    
1640
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1641
    if(!square){
1642
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1643
    }
1644

    
1645
    if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1646

    
1647
    if(MB_FIELD){
1648
        // chroma offset when predicting from a field of opposite parity
1649
        my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1650
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1651
    }
1652
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1653
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1654

    
1655
    if(emu){
1656
        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);
1657
            src_cb= s->edge_emu_buffer;
1658
    }
1659
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1660

    
1661
    if(emu){
1662
        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);
1663
            src_cr= s->edge_emu_buffer;
1664
    }
1665
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1666
}
1667

    
1668
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1669
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1670
                           int x_offset, int y_offset,
1671
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1672
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1673
                           int list0, int list1){
1674
    MpegEncContext * const s = &h->s;
1675
    qpel_mc_func *qpix_op=  qpix_put;
1676
    h264_chroma_mc_func chroma_op= chroma_put;
1677

    
1678
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1679
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1680
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1681
    x_offset += 8*s->mb_x;
1682
    y_offset += 8*(s->mb_y >> MB_FIELD);
1683

    
1684
    if(list0){
1685
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1686
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1687
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1688
                           qpix_op, chroma_op);
1689

    
1690
        qpix_op=  qpix_avg;
1691
        chroma_op= chroma_avg;
1692
    }
1693

    
1694
    if(list1){
1695
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1696
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1697
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1698
                           qpix_op, chroma_op);
1699
    }
1700
}
1701

    
1702
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1703
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1704
                           int x_offset, int y_offset,
1705
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1706
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1707
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1708
                           int list0, int list1){
1709
    MpegEncContext * const s = &h->s;
1710

    
1711
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1712
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1713
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1714
    x_offset += 8*s->mb_x;
1715
    y_offset += 8*(s->mb_y >> MB_FIELD);
1716

    
1717
    if(list0 && list1){
1718
        /* don't optimize for luma-only case, since B-frames usually
1719
         * use implicit weights => chroma too. */
1720
        uint8_t *tmp_cb = s->obmc_scratchpad;
1721
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1722
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1723
        int refn0 = h->ref_cache[0][ scan8[n] ];
1724
        int refn1 = h->ref_cache[1][ scan8[n] ];
1725

    
1726
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1727
                    dest_y, dest_cb, dest_cr,
1728
                    x_offset, y_offset, qpix_put, chroma_put);
1729
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1730
                    tmp_y, tmp_cb, tmp_cr,
1731
                    x_offset, y_offset, qpix_put, chroma_put);
1732

    
1733
        if(h->use_weight == 2){
1734
            int weight0 = h->implicit_weight[refn0][refn1];
1735
            int weight1 = 64 - weight0;
1736
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
1737
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1738
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1739
        }else{
1740
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1741
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1742
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1743
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1744
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1745
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1746
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1747
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1748
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1749
        }
1750
    }else{
1751
        int list = list1 ? 1 : 0;
1752
        int refn = h->ref_cache[list][ scan8[n] ];
1753
        Picture *ref= &h->ref_list[list][refn];
1754
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1755
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
1756
                    qpix_put, chroma_put);
1757

    
1758
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1759
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
1760
        if(h->use_weight_chroma){
1761
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1762
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1763
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1764
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1765
        }
1766
    }
1767
}
1768

    
1769
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1770
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1771
                           int x_offset, int y_offset,
1772
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1773
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1774
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1775
                           int list0, int list1){
1776
    if((h->use_weight==2 && list0 && list1
1777
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1778
       || h->use_weight==1)
1779
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1780
                         x_offset, y_offset, qpix_put, chroma_put,
1781
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1782
    else
1783
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1784
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1785
}
1786

    
1787
static inline void prefetch_motion(H264Context *h, int list){
1788
    /* fetch pixels for estimated mv 4 macroblocks ahead
1789
     * optimized for 64byte cache lines */
1790
    MpegEncContext * const s = &h->s;
1791
    const int refn = h->ref_cache[list][scan8[0]];
1792
    if(refn >= 0){
1793
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1794
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1795
        uint8_t **src= h->ref_list[list][refn].data;
1796
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1797
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
1798
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1799
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1800
    }
1801
}
1802

    
1803
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1804
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1805
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1806
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1807
    MpegEncContext * const s = &h->s;
1808
    const int mb_xy= h->mb_xy;
1809
    const int mb_type= s->current_picture.mb_type[mb_xy];
1810

    
1811
    assert(IS_INTER(mb_type));
1812

    
1813
    prefetch_motion(h, 0);
1814

    
1815
    if(IS_16X16(mb_type)){
1816
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1817
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1818
                &weight_op[0], &weight_avg[0],
1819
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1820
    }else if(IS_16X8(mb_type)){
1821
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1822
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1823
                &weight_op[1], &weight_avg[1],
1824
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1825
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1826
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1827
                &weight_op[1], &weight_avg[1],
1828
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1829
    }else if(IS_8X16(mb_type)){
1830
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1831
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1832
                &weight_op[2], &weight_avg[2],
1833
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1834
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1835
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1836
                &weight_op[2], &weight_avg[2],
1837
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1838
    }else{
1839
        int i;
1840

    
1841
        assert(IS_8X8(mb_type));
1842

    
1843
        for(i=0; i<4; i++){
1844
            const int sub_mb_type= h->sub_mb_type[i];
1845
            const int n= 4*i;
1846
            int x_offset= (i&1)<<2;
1847
            int y_offset= (i&2)<<1;
1848

    
1849
            if(IS_SUB_8X8(sub_mb_type)){
1850
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1851
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1852
                    &weight_op[3], &weight_avg[3],
1853
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1854
            }else if(IS_SUB_8X4(sub_mb_type)){
1855
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1856
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1857
                    &weight_op[4], &weight_avg[4],
1858
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1859
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1860
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1861
                    &weight_op[4], &weight_avg[4],
1862
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1863
            }else if(IS_SUB_4X8(sub_mb_type)){
1864
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1865
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1866
                    &weight_op[5], &weight_avg[5],
1867
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1868
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1869
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1870
                    &weight_op[5], &weight_avg[5],
1871
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1872
            }else{
1873
                int j;
1874
                assert(IS_SUB_4X4(sub_mb_type));
1875
                for(j=0; j<4; j++){
1876
                    int sub_x_offset= x_offset + 2*(j&1);
1877
                    int sub_y_offset= y_offset +   (j&2);
1878
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1879
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1880
                        &weight_op[6], &weight_avg[6],
1881
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1882
                }
1883
            }
1884
        }
1885
    }
1886

    
1887
    prefetch_motion(h, 1);
1888
}
1889

    
1890
static av_cold void init_cavlc_level_tab(void){
1891
    int suffix_length, mask;
1892
    unsigned int i;
1893

    
1894
    for(suffix_length=0; suffix_length<7; suffix_length++){
1895
        for(i=0; i<(1<<LEVEL_TAB_BITS); i++){
1896
            int prefix= LEVEL_TAB_BITS - av_log2(2*i);
1897
            int level_code= (prefix<<suffix_length) + (i>>(LEVEL_TAB_BITS-prefix-1-suffix_length)) - (1<<suffix_length);
1898

    
1899
            mask= -(level_code&1);
1900
            level_code= (((2+level_code)>>1) ^ mask) - mask;
1901
            if(prefix + 1 + suffix_length <= LEVEL_TAB_BITS){
1902
                cavlc_level_tab[suffix_length][i][0]= level_code;
1903
                cavlc_level_tab[suffix_length][i][1]= prefix + 1 + suffix_length;
1904
            }else if(prefix + 1 <= LEVEL_TAB_BITS){
1905
                cavlc_level_tab[suffix_length][i][0]= prefix+100;
1906
                cavlc_level_tab[suffix_length][i][1]= prefix + 1;
1907
            }else{
1908
                cavlc_level_tab[suffix_length][i][0]= LEVEL_TAB_BITS+100;
1909
                cavlc_level_tab[suffix_length][i][1]= LEVEL_TAB_BITS;
1910
            }
1911
        }
1912
    }
1913
}
1914

    
1915
static av_cold void decode_init_vlc(void){
1916
    static int done = 0;
1917

    
1918
    if (!done) {
1919
        int i;
1920
        int offset;
1921
        done = 1;
1922

    
1923
        chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
1924
        chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
1925
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1926
                 &chroma_dc_coeff_token_len [0], 1, 1,
1927
                 &chroma_dc_coeff_token_bits[0], 1, 1,
1928
                 INIT_VLC_USE_NEW_STATIC);
1929

    
1930
        offset = 0;
1931
        for(i=0; i<4; i++){
1932
            coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
1933
            coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
1934
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1935
                     &coeff_token_len [i][0], 1, 1,
1936
                     &coeff_token_bits[i][0], 1, 1,
1937
                     INIT_VLC_USE_NEW_STATIC);
1938
            offset += coeff_token_vlc_tables_size[i];
1939
        }
1940
        /*
1941
         * This is a one time safety check to make sure that
1942
         * the packed static coeff_token_vlc table sizes
1943
         * were initialized correctly.
1944
         */
1945
        assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
1946

    
1947
        for(i=0; i<3; i++){
1948
            chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
1949
            chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
1950
            init_vlc(&chroma_dc_total_zeros_vlc[i],
1951
                     CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1952
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
1953
                     &chroma_dc_total_zeros_bits[i][0], 1, 1,
1954
                     INIT_VLC_USE_NEW_STATIC);
1955
        }
1956
        for(i=0; i<15; i++){
1957
            total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
1958
            total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
1959
            init_vlc(&total_zeros_vlc[i],
1960
                     TOTAL_ZEROS_VLC_BITS, 16,
1961
                     &total_zeros_len [i][0], 1, 1,
1962
                     &total_zeros_bits[i][0], 1, 1,
1963
                     INIT_VLC_USE_NEW_STATIC);
1964
        }
1965

    
1966
        for(i=0; i<6; i++){
1967
            run_vlc[i].table = run_vlc_tables[i];
1968
            run_vlc[i].table_allocated = run_vlc_tables_size;
1969
            init_vlc(&run_vlc[i],
1970
                     RUN_VLC_BITS, 7,
1971
                     &run_len [i][0], 1, 1,
1972
                     &run_bits[i][0], 1, 1,
1973
                     INIT_VLC_USE_NEW_STATIC);
1974
        }
1975
        run7_vlc.table = run7_vlc_table,
1976
        run7_vlc.table_allocated = run7_vlc_table_size;
1977
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1978
                 &run_len [6][0], 1, 1,
1979
                 &run_bits[6][0], 1, 1,
1980
                 INIT_VLC_USE_NEW_STATIC);
1981

    
1982
        init_cavlc_level_tab();
1983
    }
1984
}
1985

    
1986
static void free_tables(H264Context *h){
1987
    int i;
1988
    H264Context *hx;
1989
    av_freep(&h->intra4x4_pred_mode);
1990
    av_freep(&h->chroma_pred_mode_table);
1991
    av_freep(&h->cbp_table);
1992
    av_freep(&h->mvd_table[0]);
1993
    av_freep(&h->mvd_table[1]);
1994
    av_freep(&h->direct_table);
1995
    av_freep(&h->non_zero_count);
1996
    av_freep(&h->slice_table_base);
1997
    h->slice_table= NULL;
1998

    
1999
    av_freep(&h->mb2b_xy);
2000
    av_freep(&h->mb2b8_xy);
2001

    
2002
    for(i = 0; i < h->s.avctx->thread_count; i++) {
2003
        hx = h->thread_context[i];
2004
        if(!hx) continue;
2005
        av_freep(&hx->top_borders[1]);
2006
        av_freep(&hx->top_borders[0]);
2007
        av_freep(&hx->s.obmc_scratchpad);
2008
    }
2009
}
2010

    
2011
static void init_dequant8_coeff_table(H264Context *h){
2012
    int i,q,x;
2013
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2014
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
2015
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
2016

    
2017
    for(i=0; i<2; i++ ){
2018
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2019
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
2020
            break;
2021
        }
2022

    
2023
        for(q=0; q<52; q++){
2024
            int shift = div6[q];
2025
            int idx = rem6[q];
2026
            for(x=0; x<64; x++)
2027
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2028
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2029
                    h->pps.scaling_matrix8[i][x]) << shift;
2030
        }
2031
    }
2032
}
2033

    
2034
static void init_dequant4_coeff_table(H264Context *h){
2035
    int i,j,q,x;
2036
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2037
    for(i=0; i<6; i++ ){
2038
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
2039
        for(j=0; j<i; j++){
2040
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2041
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2042
                break;
2043
            }
2044
        }
2045
        if(j<i)
2046
            continue;
2047

    
2048
        for(q=0; q<52; q++){
2049
            int shift = div6[q] + 2;
2050
            int idx = rem6[q];
2051
            for(x=0; x<16; x++)
2052
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2053
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2054
                    h->pps.scaling_matrix4[i][x]) << shift;
2055
        }
2056
    }
2057
}
2058

    
2059
static void init_dequant_tables(H264Context *h){
2060
    int i,x;
2061
    init_dequant4_coeff_table(h);
2062
    if(h->pps.transform_8x8_mode)
2063
        init_dequant8_coeff_table(h);
2064
    if(h->sps.transform_bypass){
2065
        for(i=0; i<6; i++)
2066
            for(x=0; x<16; x++)
2067
                h->dequant4_coeff[i][0][x] = 1<<6;
2068
        if(h->pps.transform_8x8_mode)
2069
            for(i=0; i<2; i++)
2070
                for(x=0; x<64; x++)
2071
                    h->dequant8_coeff[i][0][x] = 1<<6;
2072
    }
2073
}
2074

    
2075

    
2076
/**
2077
 * allocates tables.
2078
 * needs width/height
2079
 */
2080
static int alloc_tables(H264Context *h){
2081
    MpegEncContext * const s = &h->s;
2082
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2083
    int x,y;
2084

    
2085
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2086

    
2087
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2088
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base))
2089
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2090

    
2091
    CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2092
    CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2093
    CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2094
    CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2095

    
2096
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
2097
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2098

    
2099
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2100
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2101
    for(y=0; y<s->mb_height; y++){
2102
        for(x=0; x<s->mb_width; x++){
2103
            const int mb_xy= x + y*s->mb_stride;
2104
            const int b_xy = 4*x + 4*y*h->b_stride;
2105
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2106

    
2107
            h->mb2b_xy [mb_xy]= b_xy;
2108
            h->mb2b8_xy[mb_xy]= b8_xy;
2109
        }
2110
    }
2111

    
2112
    s->obmc_scratchpad = NULL;
2113

    
2114
    if(!h->dequant4_coeff[0])
2115
        init_dequant_tables(h);
2116

    
2117
    return 0;
2118
fail:
2119
    free_tables(h);
2120
    return -1;
2121
}
2122

    
2123
/**
2124
 * Mimic alloc_tables(), but for every context thread.
2125
 */
2126
static void clone_tables(H264Context *dst, H264Context *src){
2127
    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
2128
    dst->non_zero_count           = src->non_zero_count;
2129
    dst->slice_table              = src->slice_table;
2130
    dst->cbp_table                = src->cbp_table;
2131
    dst->mb2b_xy                  = src->mb2b_xy;
2132
    dst->mb2b8_xy                 = src->mb2b8_xy;
2133
    dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
2134
    dst->mvd_table[0]             = src->mvd_table[0];
2135
    dst->mvd_table[1]             = src->mvd_table[1];
2136
    dst->direct_table             = src->direct_table;
2137

    
2138
    dst->s.obmc_scratchpad = NULL;
2139
    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2140
}
2141

    
2142
/**
2143
 * Init context
2144
 * Allocate buffers which are not shared amongst multiple threads.
2145
 */
2146
static int context_init(H264Context *h){
2147
    CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2148
    CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2149

    
2150
    return 0;
2151
fail:
2152
    return -1; // free_tables will clean up for us
2153
}
2154

    
2155
static av_cold void common_init(H264Context *h){
2156
    MpegEncContext * const s = &h->s;
2157

    
2158
    s->width = s->avctx->width;
2159
    s->height = s->avctx->height;
2160
    s->codec_id= s->avctx->codec->id;
2161

    
2162
    ff_h264_pred_init(&h->hpc, s->codec_id);
2163

    
2164
    h->dequant_coeff_pps= -1;
2165
    s->unrestricted_mv=1;
2166
    s->decode=1; //FIXME
2167

    
2168
    dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
2169

    
2170
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2171
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2172
}
2173

    
2174
static av_cold int decode_init(AVCodecContext *avctx){
2175
    H264Context *h= avctx->priv_data;
2176
    MpegEncContext * const s = &h->s;
2177

    
2178
    MPV_decode_defaults(s);
2179

    
2180
    s->avctx = avctx;
2181
    common_init(h);
2182

    
2183
    s->out_format = FMT_H264;
2184
    s->workaround_bugs= avctx->workaround_bugs;
2185

    
2186
    // set defaults
2187
//    s->decode_mb= ff_h263_decode_mb;
2188
    s->quarter_sample = 1;
2189
    s->low_delay= 1;
2190

    
2191
    if(avctx->codec_id == CODEC_ID_SVQ3)
2192
        avctx->pix_fmt= PIX_FMT_YUVJ420P;
2193
    else if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2194
        avctx->pix_fmt= PIX_FMT_VDPAU_H264;
2195
    else
2196
        avctx->pix_fmt= PIX_FMT_YUV420P;
2197

    
2198
    decode_init_vlc();
2199

    
2200
    if(avctx->extradata_size > 0 && avctx->extradata &&
2201
       *(char *)avctx->extradata == 1){
2202
        h->is_avc = 1;
2203
        h->got_avcC = 0;
2204
    } else {
2205
        h->is_avc = 0;
2206
    }
2207

    
2208
    h->thread_context[0] = h;
2209
    h->outputed_poc = INT_MIN;
2210
    h->prev_poc_msb= 1<<16;
2211
    h->sei_recovery_frame_cnt = -1;
2212
    h->sei_dpb_output_delay = 0;
2213
    h->sei_cpb_removal_delay = -1;
2214
    h->sei_buffering_period_present = 0;
2215
    return 0;
2216
}
2217

    
2218
static int frame_start(H264Context *h){
2219
    MpegEncContext * const s = &h->s;
2220
    int i;
2221

    
2222
    if(MPV_frame_start(s, s->avctx) < 0)
2223
        return -1;
2224
    ff_er_frame_start(s);
2225
    /*
2226
     * MPV_frame_start uses pict_type to derive key_frame.
2227
     * This is incorrect for H.264; IDR markings must be used.
2228
     * Zero here; IDR markings per slice in frame or fields are ORed in later.
2229
     * See decode_nal_units().
2230
     */
2231
    s->current_picture_ptr->key_frame= 0;
2232

    
2233
    assert(s->linesize && s->uvlinesize);
2234

    
2235
    for(i=0; i<16; i++){
2236
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2237
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2238
    }
2239
    for(i=0; i<4; i++){
2240
        h->block_offset[16+i]=
2241
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2242
        h->block_offset[24+16+i]=
2243
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2244
    }
2245

    
2246
    /* can't be in alloc_tables because linesize isn't known there.
2247
     * FIXME: redo bipred weight to not require extra buffer? */
2248
    for(i = 0; i < s->avctx->thread_count; i++)
2249
        if(!h->thread_context[i]->s.obmc_scratchpad)
2250
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2251

    
2252
    /* some macroblocks will be accessed before they're available */
2253
    if(FRAME_MBAFF || s->avctx->thread_count > 1)
2254
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
2255

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

    
2258
    // We mark the current picture as non-reference after allocating it, so
2259
    // that if we break out due to an error it can be released automatically
2260
    // in the next MPV_frame_start().
2261
    // SVQ3 as well as most other codecs have only last/next/current and thus
2262
    // get released even with set reference, besides SVQ3 and others do not
2263
    // mark frames as reference later "naturally".
2264
    if(s->codec_id != CODEC_ID_SVQ3)
2265
        s->current_picture_ptr->reference= 0;
2266

    
2267
    s->current_picture_ptr->field_poc[0]=
2268
    s->current_picture_ptr->field_poc[1]= INT_MAX;
2269
    assert(s->current_picture_ptr->long_ref==0);
2270

    
2271
    return 0;
2272
}
2273

    
2274
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
2275
    MpegEncContext * const s = &h->s;
2276
    int i;
2277
    int step    = 1;
2278
    int offset  = 1;
2279
    int uvoffset= 1;
2280
    int top_idx = 1;
2281
    int skiplast= 0;
2282

    
2283
    src_y  -=   linesize;
2284
    src_cb -= uvlinesize;
2285
    src_cr -= uvlinesize;
2286

    
2287
    if(!simple && FRAME_MBAFF){
2288
        if(s->mb_y&1){
2289
            offset  = MB_MBAFF ? 1 : 17;
2290
            uvoffset= MB_MBAFF ? 1 : 9;
2291
            if(!MB_MBAFF){
2292
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y +  15*linesize);
2293
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
2294
                if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2295
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
2296
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
2297
                }
2298
            }
2299
        }else{
2300
            if(!MB_MBAFF){
2301
                h->left_border[0]= h->top_borders[0][s->mb_x][15];
2302
                if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2303
                    h->left_border[34   ]= h->top_borders[0][s->mb_x][16+7  ];
2304
                    h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
2305
                }
2306
                skiplast= 1;
2307
            }
2308
            offset  =
2309
            uvoffset=
2310
            top_idx = MB_MBAFF ? 0 : 1;
2311
        }
2312
        step= MB_MBAFF ? 2 : 1;
2313
    }
2314

    
2315
    // There are two lines saved, the line above the the top macroblock of a pair,
2316
    // and the line above the bottom macroblock
2317
    h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
2318
    for(i=1; i<17 - skiplast; i++){
2319
        h->left_border[offset+i*step]= src_y[15+i*  linesize];
2320
    }
2321

    
2322
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2323
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2324

    
2325
    if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2326
        h->left_border[uvoffset+34   ]= h->top_borders[top_idx][s->mb_x][16+7];
2327
        h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
2328
        for(i=1; i<9 - skiplast; i++){
2329
            h->left_border[uvoffset+34   +i*step]= src_cb[7+i*uvlinesize];
2330
            h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
2331
        }
2332
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2333
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2334
    }
2335
}
2336

    
2337
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){
2338
    MpegEncContext * const s = &h->s;
2339
    int temp8, i;
2340
    uint64_t temp64;
2341
    int deblock_left;
2342
    int deblock_top;
2343
    int mb_xy;
2344
    int step    = 1;
2345
    int offset  = 1;
2346
    int uvoffset= 1;
2347
    int top_idx = 1;
2348

    
2349
    if(!simple && FRAME_MBAFF){
2350
        if(s->mb_y&1){
2351
            offset  = MB_MBAFF ? 1 : 17;
2352
            uvoffset= MB_MBAFF ? 1 : 9;
2353
        }else{
2354
            offset  =
2355
            uvoffset=
2356
            top_idx = MB_MBAFF ? 0 : 1;
2357
        }
2358
        step= MB_MBAFF ? 2 : 1;
2359
    }
2360

    
2361
    if(h->deblocking_filter == 2) {
2362
        mb_xy = h->mb_xy;
2363
        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2364
        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2365
    } else {
2366
        deblock_left = (s->mb_x > 0);
2367
        deblock_top =  (s->mb_y > !!MB_FIELD);
2368
    }
2369

    
2370
    src_y  -=   linesize + 1;
2371
    src_cb -= uvlinesize + 1;
2372
    src_cr -= uvlinesize + 1;
2373

    
2374
#define XCHG(a,b,t,xchg)\
2375
t= a;\
2376
if(xchg)\
2377
    a= b;\
2378
b= t;
2379

    
2380
    if(deblock_left){
2381
        for(i = !deblock_top; i<16; i++){
2382
            XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, xchg);
2383
        }
2384
        XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, 1);
2385
    }
2386

    
2387
    if(deblock_top){
2388
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2389
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2390
        if(s->mb_x+1 < s->mb_width){
2391
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2392
        }
2393
    }
2394

    
2395
    if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2396
        if(deblock_left){
2397
            for(i = !deblock_top; i<8; i++){
2398
                XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, xchg);
2399
                XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
2400
            }
2401
            XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, 1);
2402
            XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
2403
        }
2404
        if(deblock_top){
2405
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2406
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2407
        }
2408
    }
2409
}
2410

    
2411
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2412
    MpegEncContext * const s = &h->s;
2413
    const int mb_x= s->mb_x;
2414
    const int mb_y= s->mb_y;
2415
    const int mb_xy= h->mb_xy;
2416
    const int mb_type= s->current_picture.mb_type[mb_xy];
2417
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2418
    int linesize, uvlinesize /*dct_offset*/;
2419
    int i;
2420
    int *block_offset = &h->block_offset[0];
2421
    const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2422
    /* is_h264 should always be true if SVQ3 is disabled. */
2423
    const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
2424
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2425
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2426

    
2427
    dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2428
    dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2429
    dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2430

    
2431
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2432
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2433

    
2434
    if (!simple && MB_FIELD) {
2435
        linesize   = h->mb_linesize   = s->linesize * 2;
2436
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2437
        block_offset = &h->block_offset[24];
2438
        if(mb_y&1){ //FIXME move out of this function?
2439
            dest_y -= s->linesize*15;
2440
            dest_cb-= s->uvlinesize*7;
2441
            dest_cr-= s->uvlinesize*7;
2442
        }
2443
        if(FRAME_MBAFF) {
2444
            int list;
2445
            for(list=0; list<h->list_count; list++){
2446
                if(!USES_LIST(mb_type, list))
2447
                    continue;
2448
                if(IS_16X16(mb_type)){
2449
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
2450
                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2451
                }else{
2452
                    for(i=0; i<16; i+=4){
2453
                        int ref = h->ref_cache[list][scan8[i]];
2454
                        if(ref >= 0)
2455
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2456
                    }
2457
                }
2458
            }
2459
        }
2460
    } else {
2461
        linesize   = h->mb_linesize   = s->linesize;
2462
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2463
//        dct_offset = s->linesize * 16;
2464
    }
2465

    
2466
    if (!simple && IS_INTRA_PCM(mb_type)) {
2467
        for (i=0; i<16; i++) {
2468
            memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
2469
        }
2470
        for (i=0; i<8; i++) {
2471
            memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
2472
            memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
2473
        }
2474
    } else {
2475
        if(IS_INTRA(mb_type)){
2476
            if(h->deblocking_filter)
2477
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2478

    
2479
            if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2480
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2481
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2482
            }
2483

    
2484
            if(IS_INTRA4x4(mb_type)){
2485
                if(simple || !s->encoding){
2486
                    if(IS_8x8DCT(mb_type)){
2487
                        if(transform_bypass){
2488
                            idct_dc_add =
2489
                            idct_add    = s->dsp.add_pixels8;
2490
                        }else{
2491
                            idct_dc_add = s->dsp.h264_idct8_dc_add;
2492
                            idct_add    = s->dsp.h264_idct8_add;
2493
                        }
2494
                        for(i=0; i<16; i+=4){
2495
                            uint8_t * const ptr= dest_y + block_offset[i];
2496
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2497
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2498
                                h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
2499
                            }else{
2500
                                const int nnz = h->non_zero_count_cache[ scan8[i] ];
2501
                                h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2502
                                                            (h->topright_samples_available<<i)&0x4000, linesize);
2503
                                if(nnz){
2504
                                    if(nnz == 1 && h->mb[i*16])
2505
                                        idct_dc_add(ptr, h->mb + i*16, linesize);
2506
                                    else
2507
                                        idct_add   (ptr, h->mb + i*16, linesize);
2508
                                }
2509
                            }
2510
                        }
2511
                    }else{
2512
                        if(transform_bypass){
2513
                            idct_dc_add =
2514
                            idct_add    = s->dsp.add_pixels4;
2515
                        }else{
2516
                            idct_dc_add = s->dsp.h264_idct_dc_add;
2517
                            idct_add    = s->dsp.h264_idct_add;
2518
                        }
2519
                        for(i=0; i<16; i++){
2520
                            uint8_t * const ptr= dest_y + block_offset[i];
2521
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2522

    
2523
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2524
                                h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
2525
                            }else{
2526
                                uint8_t *topright;
2527
                                int nnz, tr;
2528
                                if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2529
                                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2530
                                    assert(mb_y || linesize <= block_offset[i]);
2531
                                    if(!topright_avail){
2532
                                        tr= ptr[3 - linesize]*0x01010101;
2533
                                        topright= (uint8_t*) &tr;
2534
                                    }else
2535
                                        topright= ptr + 4 - linesize;
2536
                                }else
2537
                                    topright= NULL;
2538

    
2539
                                h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2540
                                nnz = h->non_zero_count_cache[ scan8[i] ];
2541
                                if(nnz){
2542
                                    if(is_h264){
2543
                                        if(nnz == 1 && h->mb[i*16])
2544
                                            idct_dc_add(ptr, h->mb + i*16, linesize);
2545
                                        else
2546
                                            idct_add   (ptr, h->mb + i*16, linesize);
2547
                                    }else
2548
                                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2549
                                }
2550
                            }
2551
                        }
2552
                    }
2553
                }
2554
            }else{
2555
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2556
                if(is_h264){
2557
                    if(!transform_bypass)
2558
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
2559
                }else
2560
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2561
            }
2562
            if(h->deblocking_filter)
2563
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2564
        }else if(is_h264){
2565
            hl_motion(h, dest_y, dest_cb, dest_cr,
2566
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2567
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2568
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2569
        }
2570

    
2571

    
2572
        if(!IS_INTRA4x4(mb_type)){
2573
            if(is_h264){
2574
                if(IS_INTRA16x16(mb_type)){
2575
                    if(transform_bypass){
2576
                        if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
2577
                            h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
2578
                        }else{
2579
                            for(i=0; i<16; i++){
2580
                                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2581
                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
2582
                            }
2583
                        }
2584
                    }else{
2585
                         s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2586
                    }
2587
                }else if(h->cbp&15){
2588
                    if(transform_bypass){
2589
                        const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2590
                        idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2591
                        for(i=0; i<16; i+=di){
2592
                            if(h->non_zero_count_cache[ scan8[i] ]){
2593
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2594
                            }
2595
                        }
2596
                    }else{
2597
                        if(IS_8x8DCT(mb_type)){
2598
                            s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2599
                        }else{
2600
                            s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2601
                        }
2602
                    }
2603
                }
2604
            }else{
2605
                for(i=0; i<16; i++){
2606
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2607
                        uint8_t * const ptr= dest_y + block_offset[i];
2608
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2609
                    }
2610
                }
2611
            }
2612
        }
2613

    
2614
        if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2615
            uint8_t *dest[2] = {dest_cb, dest_cr};
2616
            if(transform_bypass){
2617
                if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2618
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
2619
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
2620
                }else{
2621
                    idct_add = s->dsp.add_pixels4;
2622
                    for(i=16; i<16+8; i++){
2623
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2624
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2625
                    }
2626
                }
2627
            }else{
2628
                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]);
2629
                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]);
2630
                if(is_h264){
2631
                    idct_add = s->dsp.h264_idct_add;
2632
                    idct_dc_add = s->dsp.h264_idct_dc_add;
2633
                    for(i=16; i<16+8; i++){
2634
                        if(h->non_zero_count_cache[ scan8[i] ])
2635
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2636
                        else if(h->mb[i*16])
2637
                            idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2638
                    }
2639
                }else{
2640
                    for(i=16; i<16+8; i++){
2641
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2642
                            uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
2643
                            svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2644
                        }
2645
                    }
2646
                }
2647
            }
2648
        }
2649
    }
2650
    if(h->cbp || IS_INTRA(mb_type))
2651
        s->dsp.clear_blocks(h->mb);
2652

    
2653
    if(h->deblocking_filter) {
2654
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
2655
        fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
2656
        h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2657
        h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2658
        if (!simple && FRAME_MBAFF) {
2659
            filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2660
        } else {
2661
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2662
        }
2663
    }
2664
}
2665

    
2666
/**
2667
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2668
 */
2669
static void hl_decode_mb_simple(H264Context *h){
2670
    hl_decode_mb_internal(h, 1);
2671
}
2672

    
2673
/**
2674
 * Process a macroblock; this handles edge cases, such as interlacing.
2675
 */
2676
static void av_noinline hl_decode_mb_complex(H264Context *h){
2677
    hl_decode_mb_internal(h, 0);
2678
}
2679

    
2680
static void hl_decode_mb(H264Context *h){
2681
    MpegEncContext * const s = &h->s;
2682
    const int mb_xy= h->mb_xy;
2683
    const int mb_type= s->current_picture.mb_type[mb_xy];
2684
    int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2685

    
2686
    if (is_complex)
2687
        hl_decode_mb_complex(h);
2688
    else hl_decode_mb_simple(h);
2689
}
2690

    
2691
static void pic_as_field(Picture *pic, const int parity){
2692
    int i;
2693
    for (i = 0; i < 4; ++i) {
2694
        if (parity == PICT_BOTTOM_FIELD)
2695
            pic->data[i] += pic->linesize[i];
2696
        pic->reference = parity;
2697
        pic->linesize[i] *= 2;
2698
    }
2699
    pic->poc= pic->field_poc[parity == PICT_BOTTOM_FIELD];
2700
}
2701

    
2702
static int split_field_copy(Picture *dest, Picture *src,
2703
                            int parity, int id_add){
2704
    int match = !!(src->reference & parity);
2705

    
2706
    if (match) {
2707
        *dest = *src;
2708
        if(parity != PICT_FRAME){
2709
            pic_as_field(dest, parity);
2710
            dest->pic_id *= 2;
2711
            dest->pic_id += id_add;
2712
        }
2713
    }
2714

    
2715
    return match;
2716
}
2717

    
2718
static int build_def_list(Picture *def, Picture **in, int len, int is_long, int sel){
2719
    int i[2]={0};
2720
    int index=0;
2721

    
2722
    while(i[0]<len || i[1]<len){
2723
        while(i[0]<len && !(in[ i[0] ] && (in[ i[0] ]->reference & sel)))
2724
            i[0]++;
2725
        while(i[1]<len && !(in[ i[1] ] && (in[ i[1] ]->reference & (sel^3))))
2726
            i[1]++;
2727
        if(i[0] < len){
2728
            in[ i[0] ]->pic_id= is_long ? i[0] : in[ i[0] ]->frame_num;
2729
            split_field_copy(&def[index++], in[ i[0]++ ], sel  , 1);
2730
        }
2731
        if(i[1] < len){
2732
            in[ i[1] ]->pic_id= is_long ? i[1] : in[ i[1] ]->frame_num;
2733
            split_field_copy(&def[index++], in[ i[1]++ ], sel^3, 0);
2734
        }
2735
    }
2736

    
2737
    return index;
2738
}
2739

    
2740
static int add_sorted(Picture **sorted, Picture **src, int len, int limit, int dir){
2741
    int i, best_poc;
2742
    int out_i= 0;
2743

    
2744
    for(;;){
2745
        best_poc= dir ? INT_MIN : INT_MAX;
2746

    
2747
        for(i=0; i<len; i++){
2748
            const int poc= src[i]->poc;
2749
            if(((poc > limit) ^ dir) && ((poc < best_poc) ^ dir)){
2750
                best_poc= poc;
2751
                sorted[out_i]= src[i];
2752
            }
2753
        }
2754
        if(best_poc == (dir ? INT_MIN : INT_MAX))
2755
            break;
2756
        limit= sorted[out_i++]->poc - dir;
2757
    }
2758
    return out_i;
2759
}
2760

    
2761
/**
2762
 * fills the default_ref_list.
2763
 */
2764
static int fill_default_ref_list(H264Context *h){
2765
    MpegEncContext * const s = &h->s;
2766
    int i, len;
2767

    
2768
    if(h->slice_type_nos==FF_B_TYPE){
2769
        Picture *sorted[32];
2770
        int cur_poc, list;
2771
        int lens[2];
2772

    
2773
        if(FIELD_PICTURE)
2774
            cur_poc= s->current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
2775
        else
2776
            cur_poc= s->current_picture_ptr->poc;
2777

    
2778
        for(list= 0; list<2; list++){
2779
            len= add_sorted(sorted    , h->short_ref, h->short_ref_count, cur_poc, 1^list);
2780
            len+=add_sorted(sorted+len, h->short_ref, h->short_ref_count, cur_poc, 0^list);
2781
            assert(len<=32);
2782
            len= build_def_list(h->default_ref_list[list]    , sorted     , len, 0, s->picture_structure);
2783
            len+=build_def_list(h->default_ref_list[list]+len, h->long_ref, 16 , 1, s->picture_structure);
2784
            assert(len<=32);
2785

    
2786
            if(len < h->ref_count[list])
2787
                memset(&h->default_ref_list[list][len], 0, sizeof(Picture)*(h->ref_count[list] - len));
2788
            lens[list]= len;
2789
        }
2790

    
2791
        if(lens[0] == lens[1] && lens[1] > 1){
2792
            for(i=0; h->default_ref_list[0][i].data[0] == h->default_ref_list[1][i].data[0] && i<lens[0]; i++);
2793
            if(i == lens[0])
2794
                FFSWAP(Picture, h->default_ref_list[1][0], h->default_ref_list[1][1]);
2795
        }
2796
    }else{
2797
        len = build_def_list(h->default_ref_list[0]    , h->short_ref, h->short_ref_count, 0, s->picture_structure);
2798
        len+= build_def_list(h->default_ref_list[0]+len, h-> long_ref, 16                , 1, s->picture_structure);
2799
        assert(len <= 32);
2800
        if(len < h->ref_count[0])
2801
            memset(&h->default_ref_list[0][len], 0, sizeof(Picture)*(h->ref_count[0] - len));
2802
    }
2803
#ifdef TRACE
2804
    for (i=0; i<h->ref_count[0]; i++) {
2805
        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]);
2806
    }
2807
    if(h->slice_type_nos==FF_B_TYPE){
2808
        for (i=0; i<h->ref_count[1]; i++) {
2809
            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]);
2810
        }
2811
    }
2812
#endif
2813
    return 0;
2814
}
2815

    
2816
static void print_short_term(H264Context *h);
2817
static void print_long_term(H264Context *h);
2818

    
2819
/**
2820
 * Extract structure information about the picture described by pic_num in
2821
 * the current decoding context (frame or field). Note that pic_num is
2822
 * picture number without wrapping (so, 0<=pic_num<max_pic_num).
2823
 * @param pic_num picture number for which to extract structure information
2824
 * @param structure one of PICT_XXX describing structure of picture
2825
 *                      with pic_num
2826
 * @return frame number (short term) or long term index of picture
2827
 *         described by pic_num
2828
 */
2829
static int pic_num_extract(H264Context *h, int pic_num, int *structure){
2830
    MpegEncContext * const s = &h->s;
2831

    
2832
    *structure = s->picture_structure;
2833
    if(FIELD_PICTURE){
2834
        if (!(pic_num & 1))
2835
            /* opposite field */
2836
            *structure ^= PICT_FRAME;
2837
        pic_num >>= 1;
2838
    }
2839

    
2840
    return pic_num;
2841
}
2842

    
2843
static int decode_ref_pic_list_reordering(H264Context *h){
2844
    MpegEncContext * const s = &h->s;
2845
    int list, index, pic_structure;
2846

    
2847
    print_short_term(h);
2848
    print_long_term(h);
2849

    
2850
    for(list=0; list<h->list_count; list++){
2851
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2852

    
2853
        if(get_bits1(&s->gb)){
2854
            int pred= h->curr_pic_num;
2855

    
2856
            for(index=0; ; index++){
2857
                unsigned int reordering_of_pic_nums_idc= get_ue_golomb_31(&s->gb);
2858
                unsigned int pic_id;
2859
                int i;
2860
                Picture *ref = NULL;
2861

    
2862
                if(reordering_of_pic_nums_idc==3)
2863
                    break;
2864

    
2865
                if(index >= h->ref_count[list]){
2866
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2867
                    return -1;
2868
                }
2869

    
2870
                if(reordering_of_pic_nums_idc<3){
2871
                    if(reordering_of_pic_nums_idc<2){
2872
                        const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2873
                        int frame_num;
2874

    
2875
                        if(abs_diff_pic_num > h->max_pic_num){
2876
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2877
                            return -1;
2878
                        }
2879

    
2880
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2881
                        else                                pred+= abs_diff_pic_num;
2882
                        pred &= h->max_pic_num - 1;
2883

    
2884
                        frame_num = pic_num_extract(h, pred, &pic_structure);
2885

    
2886
                        for(i= h->short_ref_count-1; i>=0; i--){
2887
                            ref = h->short_ref[i];
2888
                            assert(ref->reference);
2889
                            assert(!ref->long_ref);
2890
                            if(
2891
                                   ref->frame_num == frame_num &&
2892
                                   (ref->reference & pic_structure)
2893
                              )
2894
                                break;
2895
                        }
2896
                        if(i>=0)
2897
                            ref->pic_id= pred;
2898
                    }else{
2899
                        int long_idx;
2900
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2901

    
2902
                        long_idx= pic_num_extract(h, pic_id, &pic_structure);
2903

    
2904
                        if(long_idx>31){
2905
                            av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
2906
                            return -1;
2907
                        }
2908
                        ref = h->long_ref[long_idx];
2909
                        assert(!(ref && !ref->reference));
2910
                        if(ref && (ref->reference & pic_structure)){
2911
                            ref->pic_id= pic_id;
2912
                            assert(ref->long_ref);
2913
                            i=0;
2914
                        }else{
2915
                            i=-1;
2916
                        }
2917
                    }
2918

    
2919
                    if (i < 0) {
2920
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2921
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2922
                    } else {
2923
                        for(i=index; i+1<h->ref_count[list]; i++){
2924
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
2925
                                break;
2926
                        }
2927
                        for(; i > index; i--){
2928
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2929
                        }
2930
                        h->ref_list[list][index]= *ref;
2931
                        if (FIELD_PICTURE){
2932
                            pic_as_field(&h->ref_list[list][index], pic_structure);
2933
                        }
2934
                    }
2935
                }else{
2936
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2937
                    return -1;
2938
                }
2939
            }
2940
        }
2941
    }
2942
    for(list=0; list<h->list_count; list++){
2943
        for(index= 0; index < h->ref_count[list]; index++){
2944
            if(!h->ref_list[list][index].data[0]){
2945
                av_log(h->s.avctx, AV_LOG_ERROR, "Missing reference picture\n");
2946
                h->ref_list[list][index]= s->current_picture; //FIXME this is not a sensible solution
2947
            }
2948
        }
2949
    }
2950

    
2951
    return 0;
2952
}
2953

    
2954
static void fill_mbaff_ref_list(H264Context *h){
2955
    int list, i, j;
2956
    for(list=0; list<2; list++){ //FIXME try list_count
2957
        for(i=0; i<h->ref_count[list]; i++){
2958
            Picture *frame = &h->ref_list[list][i];
2959
            Picture *field = &h->ref_list[list][16+2*i];
2960
            field[0] = *frame;
2961
            for(j=0; j<3; j++)
2962
                field[0].linesize[j] <<= 1;
2963
            field[0].reference = PICT_TOP_FIELD;
2964
            field[0].poc= field[0].field_poc[0];
2965
            field[1] = field[0];
2966
            for(j=0; j<3; j++)
2967
                field[1].data[j] += frame->linesize[j];
2968
            field[1].reference = PICT_BOTTOM_FIELD;
2969
            field[1].poc= field[1].field_poc[1];
2970

    
2971
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
2972
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
2973
            for(j=0; j<2; j++){
2974
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
2975
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
2976
            }
2977
        }
2978
    }
2979
    for(j=0; j<h->ref_count[1]; j++){
2980
        for(i=0; i<h->ref_count[0]; i++)
2981
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
2982
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
2983
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
2984
    }
2985
}
2986

    
2987
static int pred_weight_table(H264Context *h){
2988
    MpegEncContext * const s = &h->s;
2989
    int list, i;
2990
    int luma_def, chroma_def;
2991

    
2992
    h->use_weight= 0;
2993
    h->use_weight_chroma= 0;
2994
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2995
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2996
    luma_def = 1<<h->luma_log2_weight_denom;
2997
    chroma_def = 1<<h->chroma_log2_weight_denom;
2998

    
2999
    for(list=0; list<2; list++){
3000
        h->luma_weight_flag[list]   = 0;
3001
        h->chroma_weight_flag[list] = 0;
3002
        for(i=0; i<h->ref_count[list]; i++){
3003
            int luma_weight_flag, chroma_weight_flag;
3004

    
3005
            luma_weight_flag= get_bits1(&s->gb);
3006
            if(luma_weight_flag){
3007
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3008
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3009
                if(   h->luma_weight[list][i] != luma_def
3010
                   || h->luma_offset[list][i] != 0) {
3011
                    h->use_weight= 1;
3012
                    h->luma_weight_flag[list]= 1;
3013
                }
3014
            }else{
3015
                h->luma_weight[list][i]= luma_def;
3016
                h->luma_offset[list][i]= 0;
3017
            }
3018

    
3019
            if(CHROMA){
3020
                chroma_weight_flag= get_bits1(&s->gb);
3021
                if(chroma_weight_flag){
3022
                    int j;
3023
                    for(j=0; j<2; j++){
3024
                        h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3025
                        h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3026
                        if(   h->chroma_weight[list][i][j] != chroma_def
3027
                           || h->chroma_offset[list][i][j] != 0) {
3028
                            h->use_weight_chroma= 1;
3029
                            h->chroma_weight_flag[list]= 1;
3030
                        }
3031
                    }
3032
                }else{
3033
                    int j;
3034
                    for(j=0; j<2; j++){
3035
                        h->chroma_weight[list][i][j]= chroma_def;
3036
                        h->chroma_offset[list][i][j]= 0;
3037
                    }
3038
                }
3039
            }
3040
        }
3041
        if(h->slice_type_nos != FF_B_TYPE) break;
3042
    }
3043
    h->use_weight= h->use_weight || h->use_weight_chroma;
3044
    return 0;
3045
}
3046

    
3047
static void implicit_weight_table(H264Context *h){
3048
    MpegEncContext * const s = &h->s;
3049
    int ref0, ref1, i;
3050
    int cur_poc = s->current_picture_ptr->poc;
3051

    
3052
    for (i = 0; i < 2; i++) {
3053
        h->luma_weight_flag[i]   = 0;
3054
        h->chroma_weight_flag[i] = 0;
3055
    }
3056

    
3057
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3058
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3059
        h->use_weight= 0;
3060
        h->use_weight_chroma= 0;
3061
        return;
3062
    }
3063

    
3064
    h->use_weight= 2;
3065
    h->use_weight_chroma= 2;
3066
    h->luma_log2_weight_denom= 5;
3067
    h->chroma_log2_weight_denom= 5;
3068

    
3069
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3070
        int poc0 = h->ref_list[0][ref0].poc;
3071
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3072
            int poc1 = h->ref_list[1][ref1].poc;
3073
            int td = av_clip(poc1 - poc0, -128, 127);
3074
            if(td){
3075
                int tb = av_clip(cur_poc - poc0, -128, 127);
3076
                int tx = (16384 + (FFABS(td) >> 1)) / td;
3077
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3078
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3079
                    h->implicit_weight[ref0][ref1] = 32;
3080
                else
3081
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3082
            }else
3083
                h->implicit_weight[ref0][ref1] = 32;
3084
        }
3085
    }
3086
}
3087

    
3088
/**
3089
 * Mark a picture as no longer needed for reference. The refmask
3090
 * argument allows unreferencing of individual fields or the whole frame.
3091
 * If the picture becomes entirely unreferenced, but is being held for
3092
 * display purposes, it is marked as such.
3093
 * @param refmask mask of fields to unreference; the mask is bitwise
3094
 *                anded with the reference marking of pic
3095
 * @return non-zero if pic becomes entirely unreferenced (except possibly
3096
 *         for display purposes) zero if one of the fields remains in
3097
 *         reference
3098
 */
3099
static inline int unreference_pic(H264Context *h, Picture *pic, int refmask){
3100
    int i;
3101
    if (pic->reference &= refmask) {
3102
        return 0;
3103
    } else {
3104
        for(i = 0; h->delayed_pic[i]; i++)
3105
            if(pic == h->delayed_pic[i]){
3106
                pic->reference=DELAYED_PIC_REF;
3107
                break;
3108
            }
3109
        return 1;
3110
    }
3111
}
3112

    
3113
/**
3114
 * instantaneous decoder refresh.
3115
 */
3116
static void idr(H264Context *h){
3117
    int i;
3118

    
3119
    for(i=0; i<16; i++){
3120
        remove_long(h, i, 0);
3121
    }
3122
    assert(h->long_ref_count==0);
3123

    
3124
    for(i=0; i<h->short_ref_count; i++){
3125
        unreference_pic(h, h->short_ref[i], 0);
3126
        h->short_ref[i]= NULL;
3127
    }
3128
    h->short_ref_count=0;
3129
    h->prev_frame_num= 0;
3130
    h->prev_frame_num_offset= 0;
3131
    h->prev_poc_msb=
3132
    h->prev_poc_lsb= 0;
3133
}
3134

    
3135
/* forget old pics after a seek */
3136
static void flush_dpb(AVCodecContext *avctx){
3137
    H264Context *h= avctx->priv_data;
3138
    int i;
3139
    for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
3140
        if(h->delayed_pic[i])
3141
            h->delayed_pic[i]->reference= 0;
3142
        h->delayed_pic[i]= NULL;
3143
    }
3144
    h->outputed_poc= INT_MIN;
3145
    idr(h);
3146
    if(h->s.current_picture_ptr)
3147
        h->s.current_picture_ptr->reference= 0;
3148
    h->s.first_field= 0;
3149
    h->sei_recovery_frame_cnt = -1;
3150
    h->sei_dpb_output_delay = 0;
3151
    h->sei_cpb_removal_delay = -1;
3152
    h->sei_buffering_period_present = 0;
3153
    ff_mpeg_flush(avctx);
3154
}
3155

    
3156
/**
3157
 * Find a Picture in the short term reference list by frame number.
3158
 * @param frame_num frame number to search for
3159
 * @param idx the index into h->short_ref where returned picture is found
3160
 *            undefined if no picture found.
3161
 * @return pointer to the found picture, or NULL if no pic with the provided
3162
 *                 frame number is found
3163
 */
3164
static Picture * find_short(H264Context *h, int frame_num, int *idx){
3165
    MpegEncContext * const s = &h->s;
3166
    int i;
3167

    
3168
    for(i=0; i<h->short_ref_count; i++){
3169
        Picture *pic= h->short_ref[i];
3170
        if(s->avctx->debug&FF_DEBUG_MMCO)
3171
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3172
        if(pic->frame_num == frame_num) {
3173
            *idx = i;
3174
            return pic;
3175
        }
3176
    }
3177
    return NULL;
3178
}
3179

    
3180
/**
3181
 * Remove a picture from the short term reference list by its index in
3182
 * that list.  This does no checking on the provided index; it is assumed
3183
 * to be valid. Other list entries are shifted down.
3184
 * @param i index into h->short_ref of picture to remove.
3185
 */
3186
static void remove_short_at_index(H264Context *h, int i){
3187
    assert(i >= 0 && i < h->short_ref_count);
3188
    h->short_ref[i]= NULL;
3189
    if (--h->short_ref_count)
3190
        memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i)*sizeof(Picture*));
3191
}
3192

    
3193
/**
3194
 *
3195
 * @return the removed picture or NULL if an error occurs
3196
 */
3197
static Picture * remove_short(H264Context *h, int frame_num, int ref_mask){
3198
    MpegEncContext * const s = &h->s;
3199
    Picture *pic;
3200
    int i;
3201

    
3202
    if(s->avctx->debug&FF_DEBUG_MMCO)
3203
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3204

    
3205
    pic = find_short(h, frame_num, &i);
3206
    if (pic){
3207
        if(unreference_pic(h, pic, ref_mask))
3208
        remove_short_at_index(h, i);
3209
    }
3210

    
3211
    return pic;
3212
}
3213

    
3214
/**
3215
 * Remove a picture from the long term reference list by its index in
3216
 * that list.
3217
 * @return the removed picture or NULL if an error occurs
3218
 */
3219
static Picture * remove_long(H264Context *h, int i, int ref_mask){
3220
    Picture *pic;
3221

    
3222
    pic= h->long_ref[i];
3223
    if (pic){
3224
        if(unreference_pic(h, pic, ref_mask)){
3225
            assert(h->long_ref[i]->long_ref == 1);
3226
            h->long_ref[i]->long_ref= 0;
3227
            h->long_ref[i]= NULL;
3228
            h->long_ref_count--;
3229
        }
3230
    }
3231

    
3232
    return pic;
3233
}
3234

    
3235
/**
3236
 * print short term list
3237
 */
3238
static void print_short_term(H264Context *h) {
3239
    uint32_t i;
3240
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3241
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3242
        for(i=0; i<h->short_ref_count; i++){
3243
            Picture *pic= h->short_ref[i];
3244
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3245
        }
3246
    }
3247
}
3248

    
3249
/**
3250
 * print long term list
3251
 */
3252
static void print_long_term(H264Context *h) {
3253
    uint32_t i;
3254
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3255
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3256
        for(i = 0; i < 16; i++){
3257
            Picture *pic= h->long_ref[i];
3258
            if (pic) {
3259
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3260
            }
3261
        }
3262
    }
3263
}
3264

    
3265
/**
3266
 * Executes the reference picture marking (memory management control operations).
3267
 */
3268
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3269
    MpegEncContext * const s = &h->s;
3270
    int i, j;
3271
    int current_ref_assigned=0;
3272
    Picture *pic;
3273

    
3274
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3275
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3276

    
3277
    for(i=0; i<mmco_count; i++){
3278
        int structure, frame_num;
3279
        if(s->avctx->debug&FF_DEBUG_MMCO)
3280
            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);
3281

    
3282
        if(   mmco[i].opcode == MMCO_SHORT2UNUSED
3283
           || mmco[i].opcode == MMCO_SHORT2LONG){
3284
            frame_num = pic_num_extract(h, mmco[i].short_pic_num, &structure);
3285
            pic = find_short(h, frame_num, &j);
3286
            if(!pic){
3287
                if(mmco[i].opcode != MMCO_SHORT2LONG || !h->long_ref[mmco[i].long_arg]
3288
                   || h->long_ref[mmco[i].long_arg]->frame_num != frame_num)
3289
                av_log(h->s.avctx, AV_LOG_ERROR, "mmco: unref short failure\n");
3290
                continue;
3291
            }
3292
        }
3293

    
3294
        switch(mmco[i].opcode){
3295
        case MMCO_SHORT2UNUSED:
3296
            if(s->avctx->debug&FF_DEBUG_MMCO)
3297
                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);
3298
            remove_short(h, frame_num, structure ^ PICT_FRAME);
3299
            break;
3300
        case MMCO_SHORT2LONG:
3301
                if (h->long_ref[mmco[i].long_arg] != pic)
3302
                    remove_long(h, mmco[i].long_arg, 0);
3303

    
3304
                remove_short_at_index(h, j);
3305
                h->long_ref[ mmco[i].long_arg ]= pic;
3306
                if (h->long_ref[ mmco[i].long_arg ]){
3307
                    h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3308
                    h->long_ref_count++;
3309
                }
3310
            break;
3311
        case MMCO_LONG2UNUSED:
3312
            j = pic_num_extract(h, mmco[i].long_arg, &structure);
3313
            pic = h->long_ref[j];
3314
            if (pic) {
3315
                remove_long(h, j, structure ^ PICT_FRAME);
3316
            } else if(s->avctx->debug&FF_DEBUG_MMCO)
3317
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref long failure\n");
3318
            break;
3319
        case MMCO_LONG:
3320
                    // Comment below left from previous code as it is an interresting note.
3321
                    /* First field in pair is in short term list or
3322
                     * at a different long term index.
3323
                     * This is not allowed; see 7.4.3.3, notes 2 and 3.
3324
                     * Report the problem and keep the pair where it is,
3325
                     * and mark this field valid.
3326
                     */
3327

    
3328
            if (h->long_ref[mmco[i].long_arg] != s->current_picture_ptr) {
3329
                remove_long(h, mmco[i].long_arg, 0);
3330

    
3331
                h->long_ref[ mmco[i].long_arg ]= s->current_picture_ptr;
3332
                h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3333
                h->long_ref_count++;
3334
            }
3335

    
3336
            s->current_picture_ptr->reference |= s->picture_structure;
3337
            current_ref_assigned=1;
3338
            break;
3339
        case MMCO_SET_MAX_LONG:
3340
            assert(mmco[i].long_arg <= 16);
3341
            // just remove the long term which index is greater than new max
3342
            for(j = mmco[i].long_arg; j<16; j++){
3343
                remove_long(h, j, 0);
3344
            }
3345
            break;
3346
        case MMCO_RESET:
3347
            while(h->short_ref_count){
3348
                remove_short(h, h->short_ref[0]->frame_num, 0);
3349
            }
3350
            for(j = 0; j < 16; j++) {
3351
                remove_long(h, j, 0);
3352
            }
3353
            s->current_picture_ptr->poc=
3354
            s->current_picture_ptr->field_poc[0]=
3355
            s->current_picture_ptr->field_poc[1]=
3356
            h->poc_lsb=
3357
            h->poc_msb=
3358
            h->frame_num=
3359
            s->current_picture_ptr->frame_num= 0;
3360
            break;
3361
        default: assert(0);
3362
        }
3363
    }
3364

    
3365
    if (!current_ref_assigned) {
3366
        /* Second field of complementary field pair; the first field of
3367
         * which is already referenced. If short referenced, it
3368
         * should be first entry in short_ref. If not, it must exist
3369
         * in long_ref; trying to put it on the short list here is an
3370
         * error in the encoded bit stream (ref: 7.4.3.3, NOTE 2 and 3).
3371
         */
3372
        if (h->short_ref_count && h->short_ref[0] == s->current_picture_ptr) {
3373
            /* Just mark the second field valid */
3374
            s->current_picture_ptr->reference = PICT_FRAME;
3375
        } else if (s->current_picture_ptr->long_ref) {
3376
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term reference "
3377
                                             "assignment for second field "
3378
                                             "in complementary field pair "
3379
                                             "(first field is long term)\n");
3380
        } else {
3381
            pic= remove_short(h, s->current_picture_ptr->frame_num, 0);
3382
            if(pic){
3383
                av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3384
            }
3385

    
3386
            if(h->short_ref_count)
3387
                memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3388

    
3389
            h->short_ref[0]= s->current_picture_ptr;
3390
            h->short_ref_count++;
3391
            s->current_picture_ptr->reference |= s->picture_structure;
3392
        }
3393
    }
3394

    
3395
    if (h->long_ref_count + h->short_ref_count > h->sps.ref_frame_count){
3396

    
3397
        /* We have too many reference frames, probably due to corrupted
3398
         * stream. Need to discard one frame. Prevents overrun of the
3399
         * short_ref and long_ref buffers.
3400
         */
3401
        av_log(h->s.avctx, AV_LOG_ERROR,
3402
               "number of reference frames exceeds max (probably "
3403
               "corrupt input), discarding one\n");
3404

    
3405
        if (h->long_ref_count && !h->short_ref_count) {
3406
            for (i = 0; i < 16; ++i)
3407
                if (h->long_ref[i])
3408
                    break;
3409

    
3410
            assert(i < 16);
3411
            remove_long(h, i, 0);
3412
        } else {
3413
            pic = h->short_ref[h->short_ref_count - 1];
3414
            remove_short(h, pic->frame_num, 0);
3415
        }
3416
    }
3417

    
3418
    print_short_term(h);
3419
    print_long_term(h);
3420
    return 0;
3421
}
3422

    
3423
static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
3424
    MpegEncContext * const s = &h->s;
3425
    int i;
3426

    
3427
    h->mmco_index= 0;
3428
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3429
        s->broken_link= get_bits1(gb) -1;
3430
        if(get_bits1(gb)){
3431
            h->mmco[0].opcode= MMCO_LONG;
3432
            h->mmco[0].long_arg= 0;
3433
            h->mmco_index= 1;
3434
        }
3435
    }else{
3436
        if(get_bits1(gb)){ // adaptive_ref_pic_marking_mode_flag
3437
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
3438
                MMCOOpcode opcode= get_ue_golomb_31(gb);
3439

    
3440
                h->mmco[i].opcode= opcode;
3441
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3442
                    h->mmco[i].short_pic_num= (h->curr_pic_num - get_ue_golomb(gb) - 1) & (h->max_pic_num - 1);
3443
/*                    if(h->mmco[i].short_pic_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_pic_num ] == NULL){
3444
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
3445
                        return -1;
3446
                    }*/
3447
                }
3448
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3449
                    unsigned int long_arg= get_ue_golomb_31(gb);
3450
                    if(long_arg >= 32 || (long_arg >= 16 && !(opcode == MMCO_LONG2UNUSED && FIELD_PICTURE))){
3451
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3452
                        return -1;
3453
                    }
3454
                    h->mmco[i].long_arg= long_arg;
3455
                }
3456

    
3457
                if(opcode > (unsigned)MMCO_LONG){
3458
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3459
                    return -1;
3460
                }
3461
                if(opcode == MMCO_END)
3462
                    break;
3463
            }
3464
            h->mmco_index= i;
3465
        }else{
3466
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3467

    
3468
            if(h->short_ref_count && h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count &&
3469
                    !(FIELD_PICTURE && !s->first_field && s->current_picture_ptr->reference)) {
3470
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3471
                h->mmco[0].short_pic_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3472
                h->mmco_index= 1;
3473
                if (FIELD_PICTURE) {
3474
                    h->mmco[0].short_pic_num *= 2;
3475
                    h->mmco[1].opcode= MMCO_SHORT2UNUSED;
3476
                    h->mmco[1].short_pic_num= h->mmco[0].short_pic_num + 1;
3477
                    h->mmco_index= 2;
3478
                }
3479
            }
3480
        }
3481
    }
3482

    
3483
    return 0;
3484
}
3485

    
3486
static int init_poc(H264Context *h){
3487
    MpegEncContext * const s = &h->s;
3488
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3489
    int field_poc[2];
3490
    Picture *cur = s->current_picture_ptr;
3491

    
3492
    h->frame_num_offset= h->prev_frame_num_offset;
3493
    if(h->frame_num < h->prev_frame_num)
3494
        h->frame_num_offset += max_frame_num;
3495

    
3496
    if(h->sps.poc_type==0){
3497
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3498

    
3499
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3500
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3501
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3502
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3503
        else
3504
            h->poc_msb = h->prev_poc_msb;
3505
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3506
        field_poc[0] =
3507
        field_poc[1] = h->poc_msb + h->poc_lsb;
3508
        if(s->picture_structure == PICT_FRAME)
3509
            field_poc[1] += h->delta_poc_bottom;
3510
    }else if(h->sps.poc_type==1){
3511
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3512
        int i;
3513

    
3514
        if(h->sps.poc_cycle_length != 0)
3515
            abs_frame_num = h->frame_num_offset + h->frame_num;
3516
        else
3517
            abs_frame_num = 0;
3518

    
3519
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
3520
            abs_frame_num--;
3521

    
3522
        expected_delta_per_poc_cycle = 0;
3523
        for(i=0; i < h->sps.poc_cycle_length; i++)
3524
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3525

    
3526
        if(abs_frame_num > 0){
3527
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3528
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3529

    
3530
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3531
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
3532
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3533
        } else
3534
            expectedpoc = 0;
3535

    
3536
        if(h->nal_ref_idc == 0)
3537
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3538

    
3539
        field_poc[0] = expectedpoc + h->delta_poc[0];
3540
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3541

    
3542
        if(s->picture_structure == PICT_FRAME)
3543
            field_poc[1] += h->delta_poc[1];
3544
    }else{
3545
        int poc= 2*(h->frame_num_offset + h->frame_num);
3546

    
3547
        if(!h->nal_ref_idc)
3548
            poc--;
3549

    
3550
        field_poc[0]= poc;
3551
        field_poc[1]= poc;
3552
    }
3553

    
3554
    if(s->picture_structure != PICT_BOTTOM_FIELD)
3555
        s->current_picture_ptr->field_poc[0]= field_poc[0];
3556
    if(s->picture_structure != PICT_TOP_FIELD)
3557
        s->current_picture_ptr->field_poc[1]= field_poc[1];
3558
    cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
3559

    
3560
    return 0;
3561
}
3562

    
3563

    
3564
/**
3565
 * initialize scan tables
3566
 */
3567
static void init_scan_tables(H264Context *h){
3568
    MpegEncContext * const s = &h->s;
3569
    int i;
3570
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
3571
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
3572
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
3573
    }else{
3574
        for(i=0; i<16; i++){
3575
#define T(x) (x>>2) | ((x<<2) & 0xF)
3576
            h->zigzag_scan[i] = T(zigzag_scan[i]);
3577
            h-> field_scan[i] = T( field_scan[i]);
3578
#undef T
3579
        }
3580
    }
3581
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
3582
        memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
3583
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
3584
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
3585
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
3586
    }else{
3587
        for(i=0; i<64; i++){
3588
#define T(x) (x>>3) | ((x&7)<<3)
3589
            h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
3590
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
3591
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
3592
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
3593
#undef T
3594
        }
3595
    }
3596
    if(h->sps.transform_bypass){ //FIXME same ugly
3597
        h->zigzag_scan_q0          = zigzag_scan;
3598
        h->zigzag_scan8x8_q0       = ff_zigzag_direct;
3599
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
3600
        h->field_scan_q0           = field_scan;
3601
        h->field_scan8x8_q0        = field_scan8x8;
3602
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
3603
    }else{
3604
        h->zigzag_scan_q0          = h->zigzag_scan;
3605
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
3606
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
3607
        h->field_scan_q0           = h->field_scan;
3608
        h->field_scan8x8_q0        = h->field_scan8x8;
3609
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
3610
    }
3611
}
3612

    
3613
/**
3614
 * Replicates H264 "master" context to thread contexts.
3615
 */
3616
static void clone_slice(H264Context *dst, H264Context *src)
3617
{
3618
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
3619
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
3620
    dst->s.current_picture      = src->s.current_picture;
3621
    dst->s.linesize             = src->s.linesize;
3622
    dst->s.uvlinesize           = src->s.uvlinesize;
3623
    dst->s.first_field          = src->s.first_field;
3624

    
3625
    dst->prev_poc_msb           = src->prev_poc_msb;
3626
    dst->prev_poc_lsb           = src->prev_poc_lsb;
3627
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
3628
    dst->prev_frame_num         = src->prev_frame_num;
3629
    dst->short_ref_count        = src->short_ref_count;
3630

    
3631
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
3632
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
3633
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
3634
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
3635

    
3636
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
3637
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
3638
}
3639

    
3640
/**
3641
 * decodes a slice header.
3642
 * This will also call MPV_common_init() and frame_start() as needed.
3643
 *
3644
 * @param h h264context
3645
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
3646
 *
3647
 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3648
 */
3649
static int decode_slice_header(H264Context *h, H264Context *h0){
3650
    MpegEncContext * const s = &h->s;
3651
    MpegEncContext * const s0 = &h0->s;
3652
    unsigned int first_mb_in_slice;
3653
    unsigned int pps_id;
3654
    int num_ref_idx_active_override_flag;
3655
    unsigned int slice_type, tmp, i, j;
3656
    int default_ref_list_done = 0;
3657
    int last_pic_structure;
3658

    
3659
    s->dropable= h->nal_ref_idc == 0;
3660

    
3661
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
3662
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
3663
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
3664
    }else{
3665
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
3666
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
3667
    }
3668

    
3669
    first_mb_in_slice= get_ue_golomb(&s->gb);
3670

    
3671
    if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
3672
        h0->current_slice = 0;
3673
        if (!s0->first_field)
3674
            s->current_picture_ptr= NULL;
3675
    }
3676

    
3677
    slice_type= get_ue_golomb_31(&s->gb);
3678
    if(slice_type > 9){
3679
        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);
3680
        return -1;
3681
    }
3682
    if(slice_type > 4){
3683
        slice_type -= 5;
3684
        h->slice_type_fixed=1;
3685
    }else
3686
        h->slice_type_fixed=0;
3687

    
3688
    slice_type= golomb_to_pict_type[ slice_type ];
3689
    if (slice_type == FF_I_TYPE
3690
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
3691
        default_ref_list_done = 1;
3692
    }
3693
    h->slice_type= slice_type;
3694
    h->slice_type_nos= slice_type & 3;
3695

    
3696
    s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
3697
    if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
3698
        av_log(h->s.avctx, AV_LOG_ERROR,
3699
               "B picture before any references, skipping\n");
3700
        return -1;
3701
    }
3702

    
3703
    pps_id= get_ue_golomb(&s->gb);
3704
    if(pps_id>=MAX_PPS_COUNT){
3705
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3706
        return -1;
3707
    }
3708
    if(!h0->pps_buffers[pps_id]) {
3709
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS referenced\n");
3710
        return -1;
3711
    }
3712
    h->pps= *h0->pps_buffers[pps_id];
3713

    
3714
    if(!h0->sps_buffers[h->pps.sps_id]) {
3715
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS referenced\n");
3716
        return -1;
3717
    }
3718
    h->sps = *h0->sps_buffers[h->pps.sps_id];
3719

    
3720
    if(h == h0 && h->dequant_coeff_pps != pps_id){
3721
        h->dequant_coeff_pps = pps_id;
3722
        init_dequant_tables(h);
3723
    }
3724

    
3725
    s->mb_width= h->sps.mb_width;
3726
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3727

    
3728
    h->b_stride=  s->mb_width*4;
3729
    h->b8_stride= s->mb_width*2;
3730

    
3731
    s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
3732
    if(h->sps.frame_mbs_only_flag)
3733
        s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
3734
    else
3735
        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
3736

    
3737
    if (s->context_initialized
3738
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3739
        if(h != h0)
3740
            return -1;   // width / height changed during parallelized decoding
3741
        free_tables(h);
3742
        flush_dpb(s->avctx);
3743
        MPV_common_end(s);
3744
    }
3745
    if (!s->context_initialized) {
3746
        if(h != h0)
3747
            return -1;  // we cant (re-)initialize context during parallel decoding
3748
        if (MPV_common_init(s) < 0)
3749
            return -1;
3750
        s->first_field = 0;
3751

    
3752
        init_scan_tables(h);
3753
        alloc_tables(h);
3754

    
3755
        for(i = 1; i < s->avctx->thread_count; i++) {
3756
            H264Context *c;
3757
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
3758
            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
3759
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
3760
            c->sps = h->sps;
3761
            c->pps = h->pps;
3762
            init_scan_tables(c);
3763
            clone_tables(c, h);
3764
        }
3765

    
3766
        for(i = 0; i < s->avctx->thread_count; i++)
3767
            if(context_init(h->thread_context[i]) < 0)
3768
                return -1;
3769

    
3770
        s->avctx->width = s->width;
3771
        s->avctx->height = s->height;
3772
        s->avctx->sample_aspect_ratio= h->sps.sar;
3773
        if(!s->avctx->sample_aspect_ratio.den)
3774
            s->avctx->sample_aspect_ratio.den = 1;
3775

    
3776
        if(h->sps.timing_info_present_flag){
3777
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale};
3778
            if(h->x264_build > 0 && h->x264_build < 44)
3779
                s->avctx->time_base.den *= 2;
3780
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3781
                      s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
3782
        }
3783
    }
3784

    
3785
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3786

    
3787
    h->mb_mbaff = 0;
3788
    h->mb_aff_frame = 0;
3789
    last_pic_structure = s0->picture_structure;
3790
    if(h->sps.frame_mbs_only_flag){
3791
        s->picture_structure= PICT_FRAME;
3792
    }else{
3793
        if(get_bits1(&s->gb)) { //field_pic_flag
3794
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3795
        } else {
3796
            s->picture_structure= PICT_FRAME;
3797
            h->mb_aff_frame = h->sps.mb_aff;
3798
        }
3799
    }
3800
    h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
3801

    
3802
    if(h0->current_slice == 0){
3803
        while(h->frame_num !=  h->prev_frame_num &&
3804
              h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
3805
            av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
3806
            frame_start(h);
3807
            h->prev_frame_num++;
3808
            h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
3809
            s->current_picture_ptr->frame_num= h->prev_frame_num;
3810
            execute_ref_pic_marking(h, NULL, 0);
3811
        }
3812

    
3813
        /* See if we have a decoded first field looking for a pair... */
3814
        if (s0->first_field) {
3815
            assert(s0->current_picture_ptr);
3816
            assert(s0->current_picture_ptr->data[0]);
3817
            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
3818

    
3819
            /* figure out if we have a complementary field pair */
3820
            if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
3821
                /*
3822
                 * Previous field is unmatched. Don't display it, but let it
3823
                 * remain for reference if marked as such.
3824
                 */
3825
                s0->current_picture_ptr = NULL;
3826
                s0->first_field = FIELD_PICTURE;
3827

    
3828
            } else {
3829
                if (h->nal_ref_idc &&
3830
                        s0->current_picture_ptr->reference &&
3831
                        s0->current_picture_ptr->frame_num != h->frame_num) {
3832
                    /*
3833
                     * This and previous field were reference, but had
3834
                     * different frame_nums. Consider this field first in
3835
                     * pair. Throw away previous field except for reference
3836
                     * purposes.
3837
                     */
3838
                    s0->first_field = 1;
3839
                    s0->current_picture_ptr = NULL;
3840

    
3841
                } else {
3842
                    /* Second field in complementary pair */
3843
                    s0->first_field = 0;
3844
                }
3845
            }
3846

    
3847
        } else {
3848
            /* Frame or first field in a potentially complementary pair */
3849
            assert(!s0->current_picture_ptr);
3850
            s0->first_field = FIELD_PICTURE;
3851
        }
3852

    
3853
        if((!FIELD_PICTURE || s0->first_field) && frame_start(h) < 0) {
3854
            s0->first_field = 0;
3855
            return -1;
3856
        }
3857
    }
3858
    if(h != h0)
3859
        clone_slice(h, h0);
3860

    
3861
    s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
3862

    
3863
    assert(s->mb_num == s->mb_width * s->mb_height);
3864
    if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
3865
       first_mb_in_slice                    >= s->mb_num){
3866
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3867
        return -1;
3868
    }
3869
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3870
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
3871
    if (s->picture_structure == PICT_BOTTOM_FIELD)
3872
        s->resync_mb_y = s->mb_y = s->mb_y + 1;
3873
    assert(s->mb_y < s->mb_height);
3874

    
3875
    if(s->picture_structure==PICT_FRAME){
3876
        h->curr_pic_num=   h->frame_num;
3877
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3878
    }else{
3879
        h->curr_pic_num= 2*h->frame_num + 1;
3880
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3881
    }
3882

    
3883
    if(h->nal_unit_type == NAL_IDR_SLICE){
3884
        get_ue_golomb(&s->gb); /* idr_pic_id */
3885
    }
3886

    
3887
    if(h->sps.poc_type==0){
3888
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3889

    
3890
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3891
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3892
        }
3893
    }
3894

    
3895
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3896
        h->delta_poc[0]= get_se_golomb(&s->gb);
3897

    
3898
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3899
            h->delta_poc[1]= get_se_golomb(&s->gb);
3900
    }
3901

    
3902
    init_poc(h);
3903

    
3904
    if(h->pps.redundant_pic_cnt_present){
3905
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3906
    }
3907

    
3908
    //set defaults, might be overridden a few lines later
3909
    h->ref_count[0]= h->pps.ref_count[0];
3910
    h->ref_count[1]= h->pps.ref_count[1];
3911

    
3912
    if(h->slice_type_nos != FF_I_TYPE){
3913
        if(h->slice_type_nos == FF_B_TYPE){
3914
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3915
        }
3916
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3917

    
3918
        if(num_ref_idx_active_override_flag){
3919
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3920
            if(h->slice_type_nos==FF_B_TYPE)
3921
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3922

    
3923
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
3924
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3925
                h->ref_count[0]= h->ref_count[1]= 1;
3926
                return -1;
3927
            }
3928
        }
3929
        if(h->slice_type_nos == FF_B_TYPE)
3930
            h->list_count= 2;
3931
        else
3932
            h->list_count= 1;
3933
    }else
3934
        h->list_count= 0;
3935

    
3936
    if(!default_ref_list_done){
3937
        fill_default_ref_list(h);
3938
    }
3939

    
3940
    if(h->slice_type_nos!=FF_I_TYPE && decode_ref_pic_list_reordering(h) < 0)
3941
        return -1;
3942

    
3943
    if(h->slice_type_nos!=FF_I_TYPE){
3944
        s->last_picture_ptr= &h->ref_list[0][0];
3945
        ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3946
    }
3947
    if(h->slice_type_nos==FF_B_TYPE){
3948
        s->next_picture_ptr= &h->ref_list[1][0];
3949
        ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3950
    }
3951

    
3952
    if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
3953
       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
3954
        pred_weight_table(h);
3955
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
3956
        implicit_weight_table(h);
3957
    else {
3958
        h->use_weight = 0;
3959
        for (i = 0; i < 2; i++) {
3960
            h->luma_weight_flag[i]   = 0;
3961
            h->chroma_weight_flag[i] = 0;
3962
        }
3963
    }
3964

    
3965
    if(h->nal_ref_idc)
3966
        decode_ref_pic_marking(h0, &s->gb);
3967

    
3968
    if(FRAME_MBAFF)
3969
        fill_mbaff_ref_list(h);
3970

    
3971
    if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
3972
        direct_dist_scale_factor(h);
3973
    direct_ref_list_init(h);
3974

    
3975
    if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
3976
        tmp = get_ue_golomb_31(&s->gb);
3977
        if(tmp > 2){
3978
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3979
            return -1;
3980
        }
3981
        h->cabac_init_idc= tmp;
3982
    }
3983

    
3984
    h->last_qscale_diff = 0;
3985
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3986
    if(tmp>51){
3987
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3988
        return -1;
3989
    }
3990
    s->qscale= tmp;
3991
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3992
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3993
    //FIXME qscale / qp ... stuff
3994
    if(h->slice_type == FF_SP_TYPE){
3995
        get_bits1(&s->gb); /* sp_for_switch_flag */
3996
    }
3997
    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
3998
        get_se_golomb(&s->gb); /* slice_qs_delta */
3999
    }
4000

    
4001
    h->deblocking_filter = 1;
4002
    h->slice_alpha_c0_offset = 0;
4003
    h->slice_beta_offset = 0;
4004
    if( h->pps.deblocking_filter_parameters_present ) {
4005
        tmp= get_ue_golomb_31(&s->gb);
4006
        if(tmp > 2){
4007
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
4008
            return -1;
4009
        }
4010
        h->deblocking_filter= tmp;
4011
        if(h->deblocking_filter < 2)
4012
            h->deblocking_filter^= 1; // 1<->0
4013

    
4014
        if( h->deblocking_filter ) {
4015
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
4016
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
4017
        }
4018
    }
4019

    
4020
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
4021
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
4022
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
4023
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
4024
        h->deblocking_filter= 0;
4025

    
4026
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
4027
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
4028
            /* Cheat slightly for speed:
4029
               Do not bother to deblock across slices. */
4030
            h->deblocking_filter = 2;
4031
        } else {
4032
            h0->max_contexts = 1;
4033
            if(!h0->single_decode_warning) {
4034
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
4035
                h0->single_decode_warning = 1;
4036
            }
4037
            if(h != h0)
4038
                return 1; // deblocking switched inside frame
4039
        }
4040
    }
4041

    
4042
#if 0 //FMO
4043
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
4044
        slice_group_change_cycle= get_bits(&s->gb, ?);
4045
#endif
4046

    
4047
    h0->last_slice_type = slice_type;
4048
    h->slice_num = ++h0->current_slice;
4049
    if(h->slice_num >= MAX_SLICES){
4050
        av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
4051
    }
4052

    
4053
    for(j=0; j<2; j++){
4054
        int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
4055
        ref2frm[0]=
4056
        ref2frm[1]= -1;
4057
        for(i=0; i<16; i++)
4058
            ref2frm[i+2]= 4*h->ref_list[j][i].frame_num
4059
                          +(h->ref_list[j][i].reference&3);
4060
        ref2frm[18+0]=
4061
        ref2frm[18+1]= -1;
4062
        for(i=16; i<48; i++)
4063
            ref2frm[i+4]= 4*h->ref_list[j][i].frame_num
4064
                          +(h->ref_list[j][i].reference&3);
4065
    }
4066

    
4067
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
4068
    h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
4069

    
4070
    s->avctx->refs= h->sps.ref_frame_count;
4071

    
4072
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4073
        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",
4074
               h->slice_num,
4075
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
4076
               first_mb_in_slice,
4077
               av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
4078
               pps_id, h->frame_num,
4079
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4080
               h->ref_count[0], h->ref_count[1],
4081
               s->qscale,
4082
               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4083
               h->use_weight,
4084
               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
4085
               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
4086
               );
4087
    }
4088

    
4089
    return 0;
4090
}
4091

    
4092
/**
4093
 *
4094
 */
4095
static inline int get_level_prefix(GetBitContext *gb){
4096
    unsigned int buf;
4097
    int log;
4098

    
4099
    OPEN_READER(re, gb);
4100
    UPDATE_CACHE(re, gb);
4101
    buf=GET_CACHE(re, gb);
4102

    
4103
    log= 32 - av_log2(buf);
4104
#ifdef TRACE
4105
    print_bin(buf>>(32-log), log);
4106
    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__);
4107
#endif
4108

    
4109
    LAST_SKIP_BITS(re, gb, log);
4110
    CLOSE_READER(re, gb);
4111

    
4112
    return log-1;
4113
}
4114

    
4115
static inline int get_dct8x8_allowed(H264Context *h){
4116
    if(h->sps.direct_8x8_inference_flag)
4117
        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
4118
    else
4119
        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
4120
}
4121

    
4122
/**
4123
 * decodes a residual block.
4124
 * @param n block index
4125
 * @param scantable scantable
4126
 * @param max_coeff number of coefficients in the block
4127
 * @return <0 if an error occurred
4128
 */
4129
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
4130
    MpegEncContext * const s = &h->s;
4131
    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};
4132
    int level[16];
4133
    int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;
4134

    
4135
    //FIXME put trailing_onex into the context
4136

    
4137
    if(n == CHROMA_DC_BLOCK_INDEX){
4138
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4139
        total_coeff= coeff_token>>2;
4140
    }else{
4141
        if(n == LUMA_DC_BLOCK_INDEX){
4142
            total_coeff= pred_non_zero_count(h, 0);
4143
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4144
            total_coeff= coeff_token>>2;
4145
        }else{
4146
            total_coeff= pred_non_zero_count(h, n);
4147
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4148
            total_coeff= coeff_token>>2;
4149
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4150
        }
4151
    }
4152

    
4153
    //FIXME set last_non_zero?
4154

    
4155
    if(total_coeff==0)
4156
        return 0;
4157
    if(total_coeff > (unsigned)max_coeff) {
4158
        av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
4159
        return -1;
4160
    }
4161

    
4162
    trailing_ones= coeff_token&3;
4163
    tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
4164
    assert(total_coeff<=16);
4165

    
4166
    i = show_bits(gb, 3);
4167
    skip_bits(gb, trailing_ones);
4168
    level[0] = 1-((i&4)>>1);
4169
    level[1] = 1-((i&2)   );
4170
    level[2] = 1-((i&1)<<1);
4171

    
4172
    if(trailing_ones<total_coeff) {
4173
        int mask, prefix;
4174
        int suffix_length = total_coeff > 10 && trailing_ones < 3;
4175
        int bitsi= show_bits(gb, LEVEL_TAB_BITS);
4176
        int level_code= cavlc_level_tab[suffix_length][bitsi][0];
4177

    
4178
        skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
4179
        if(level_code >= 100){
4180
            prefix= level_code - 100;
4181
            if(prefix == LEVEL_TAB_BITS)
4182
                prefix += get_level_prefix(gb);
4183

    
4184
            //first coefficient has suffix_length equal to 0 or 1
4185
            if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4186
                if(suffix_length)
4187
                    level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4188
                else
4189
                    level_code= (prefix<<suffix_length); //part
4190
            }else if(prefix==14){
4191
                if(suffix_length)
4192
                    level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4193
                else
4194
                    level_code= prefix + get_bits(gb, 4); //part
4195
            }else{
4196
                level_code= (15<<suffix_length) + get_bits(gb, prefix-3); //part
4197
                if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
4198
                if(prefix>=16)
4199
                    level_code += (1<<(prefix-3))-4096;
4200
            }
4201

    
4202
            if(trailing_ones < 3) level_code += 2;
4203

    
4204
            suffix_length = 2;
4205
            mask= -(level_code&1);
4206
            level[trailing_ones]= (((2+level_code)>>1) ^ mask) - mask;
4207
        }else{
4208
            if(trailing_ones < 3) level_code += (level_code>>31)|1;
4209

    
4210
            suffix_length = 1;
4211
            if(level_code + 3U > 6U)
4212
                suffix_length++;
4213
            level[trailing_ones]= level_code;
4214
        }
4215

    
4216
        //remaining coefficients have suffix_length > 0
4217
        for(i=trailing_ones+1;i<total_coeff;i++) {
4218
            static const unsigned int suffix_limit[7] = {0,3,6,12,24,48,INT_MAX };
4219
            int bitsi= show_bits(gb, LEVEL_TAB_BITS);
4220
            level_code= cavlc_level_tab[suffix_length][bitsi][0];
4221

    
4222
            skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
4223
            if(level_code >= 100){
4224
                prefix= level_code - 100;
4225
                if(prefix == LEVEL_TAB_BITS){
4226
                    prefix += get_level_prefix(gb);
4227
                }
4228
                if(prefix<15){
4229
                    level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
4230
                }else{
4231
                    level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
4232
                    if(prefix>=16)
4233
                        level_code += (1<<(prefix-3))-4096;
4234
                }
4235
                mask= -(level_code&1);
4236
                level_code= (((2+level_code)>>1) ^ mask) - mask;
4237
            }
4238
            level[i]= level_code;
4239

    
4240
            if(suffix_limit[suffix_length] + level_code > 2U*suffix_limit[suffix_length])
4241
                suffix_length++;
4242
        }
4243
    }
4244

    
4245
    if(total_coeff == max_coeff)
4246
        zeros_left=0;
4247
    else{
4248
        if(n == CHROMA_DC_BLOCK_INDEX)
4249
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4250
        else
4251
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4252
    }
4253

    
4254
    coeff_num = zeros_left + total_coeff - 1;
4255
    j = scantable[coeff_num];
4256
    if(n > 24){
4257
        block[j] = level[0];
4258
        for(i=1;i<total_coeff;i++) {
4259
            if(zeros_left <= 0)
4260
                run_before = 0;
4261
            else if(zeros_left < 7){
4262
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4263
            }else{
4264
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4265
            }
4266
            zeros_left -= run_before;
4267
            coeff_num -= 1 + run_before;
4268
            j= scantable[ coeff_num ];
4269

    
4270
            block[j]= level[i];
4271
        }
4272
    }else{
4273
        block[j] = (level[0] * qmul[j] + 32)>>6;
4274
        for(i=1;i<total_coeff;i++) {
4275
            if(zeros_left <= 0)
4276
                run_before = 0;
4277
            else if(zeros_left < 7){
4278
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4279
            }else{
4280
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4281
            }
4282
            zeros_left -= run_before;
4283
            coeff_num -= 1 + run_before;
4284
            j= scantable[ coeff_num ];
4285

    
4286
            block[j]= (level[i] * qmul[j] + 32)>>6;
4287
        }
4288
    }
4289

    
4290
    if(zeros_left<0){
4291
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4292
        return -1;
4293
    }
4294

    
4295
    return 0;
4296
}
4297

    
4298
static void predict_field_decoding_flag(H264Context *h){
4299
    MpegEncContext * const s = &h->s;
4300
    const int mb_xy= h->mb_xy;
4301
    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
4302
                ? s->current_picture.mb_type[mb_xy-1]
4303
                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
4304
                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
4305
                : 0;
4306
    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4307
}
4308

    
4309
/**
4310
 * decodes a P_SKIP or B_SKIP macroblock
4311
 */
4312
static void decode_mb_skip(H264Context *h){
4313
    MpegEncContext * const s = &h->s;
4314
    const int mb_xy= h->mb_xy;
4315
    int mb_type=0;
4316

    
4317
    memset(h->non_zero_count[mb_xy], 0, 16);
4318
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4319

    
4320
    if(MB_FIELD)
4321
        mb_type|= MB_TYPE_INTERLACED;
4322

    
4323
    if( h->slice_type_nos == FF_B_TYPE )
4324
    {
4325
        // just for fill_caches. pred_direct_motion will set the real mb_type
4326
        mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4327

    
4328
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4329
        pred_direct_motion(h, &mb_type);
4330
        mb_type|= MB_TYPE_SKIP;
4331
    }
4332
    else
4333
    {
4334
        int mx, my;
4335
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4336

    
4337
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4338
        pred_pskip_motion(h, &mx, &my);
4339
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4340
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4341
    }
4342

    
4343
    write_back_motion(h, mb_type);
4344
    s->current_picture.mb_type[mb_xy]= mb_type;
4345
    s->current_picture.qscale_table[mb_xy]= s->qscale;
4346
    h->slice_table[ mb_xy ]= h->slice_num;
4347
    h->prev_mb_skipped= 1;
4348
}
4349

    
4350
/**
4351
 * decodes a macroblock
4352
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4353
 */
4354
static int decode_mb_cavlc(H264Context *h){
4355
    MpegEncContext * const s = &h->s;
4356
    int mb_xy;
4357
    int partition_count;
4358
    unsigned int mb_type, cbp;
4359
    int dct8x8_allowed= h->pps.transform_8x8_mode;
4360

    
4361
    mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4362

    
4363
    tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4364
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4365
                down the code */
4366
    if(h->slice_type_nos != FF_I_TYPE){
4367
        if(s->mb_skip_run==-1)
4368
            s->mb_skip_run= get_ue_golomb(&s->gb);
4369

    
4370
        if (s->mb_skip_run--) {
4371
            if(FRAME_MBAFF && (s->mb_y&1) == 0){
4372
                if(s->mb_skip_run==0)
4373
                    h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4374
                else
4375
                    predict_field_decoding_flag(h);
4376
            }
4377
            decode_mb_skip(h);
4378
            return 0;
4379
        }
4380
    }
4381
    if(FRAME_MBAFF){
4382
        if( (s->mb_y&1) == 0 )
4383
            h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4384
    }
4385

    
4386
    h->prev_mb_skipped= 0;
4387

    
4388
    mb_type= get_ue_golomb(&s->gb);
4389
    if(h->slice_type_nos == FF_B_TYPE){
4390
        if(mb_type < 23){
4391
            partition_count= b_mb_type_info[mb_type].partition_count;
4392
            mb_type=         b_mb_type_info[mb_type].type;
4393
        }else{
4394
            mb_type -= 23;
4395
            goto decode_intra_mb;
4396
        }
4397
    }else if(h->slice_type_nos == FF_P_TYPE){
4398
        if(mb_type < 5){
4399
            partition_count= p_mb_type_info[mb_type].partition_count;
4400
            mb_type=         p_mb_type_info[mb_type].type;
4401
        }else{
4402
            mb_type -= 5;
4403
            goto decode_intra_mb;
4404
        }
4405
    }else{
4406
       assert(h->slice_type_nos == FF_I_TYPE);
4407
        if(h->slice_type == FF_SI_TYPE && mb_type)
4408
            mb_type--;
4409
decode_intra_mb:
4410
        if(mb_type > 25){
4411
            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);
4412
            return -1;
4413
        }
4414
        partition_count=0;
4415
        cbp= i_mb_type_info[mb_type].cbp;
4416
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4417
        mb_type= i_mb_type_info[mb_type].type;
4418
    }
4419

    
4420
    if(MB_FIELD)
4421
        mb_type |= MB_TYPE_INTERLACED;
4422

    
4423
    h->slice_table[ mb_xy ]= h->slice_num;
4424

    
4425
    if(IS_INTRA_PCM(mb_type)){
4426
        unsigned int x;
4427

    
4428
        // We assume these blocks are very rare so we do not optimize it.
4429
        align_get_bits(&s->gb);
4430

    
4431
        // The pixels are stored in the same order as levels in h->mb array.
4432
        for(x=0; x < (CHROMA ? 384 : 256); x++){
4433
            ((uint8_t*)h->mb)[x]= get_bits(&s->gb, 8);
4434
        }
4435

    
4436
        // In deblocking, the quantizer is 0
4437
        s->current_picture.qscale_table[mb_xy]= 0;
4438
        // All coeffs are present
4439
        memset(h->non_zero_count[mb_xy], 16, 16);
4440

    
4441
        s->current_picture.mb_type[mb_xy]= mb_type;
4442
        return 0;
4443
    }
4444

    
4445
    if(MB_MBAFF){
4446
        h->ref_count[0] <<= 1;
4447
        h->ref_count[1] <<= 1;
4448
    }
4449

    
4450
    fill_caches(h, mb_type, 0);
4451

    
4452
    //mb_pred
4453
    if(IS_INTRA(mb_type)){
4454
        int pred_mode;
4455
//            init_top_left_availability(h);
4456
        if(IS_INTRA4x4(mb_type)){
4457
            int i;
4458
            int di = 1;
4459
            if(dct8x8_allowed && get_bits1(&s->gb)){
4460
                mb_type |= MB_TYPE_8x8DCT;
4461
                di = 4;
4462
            }
4463

    
4464
//                fill_intra4x4_pred_table(h);
4465
            for(i=0; i<16; i+=di){
4466
                int mode= pred_intra_mode(h, i);
4467

    
4468
                if(!get_bits1(&s->gb)){
4469