Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 53c193a9

History | View | Annotate | Download (305 KB)

1
/*
2
 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3
 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21

    
22
/**
23
 * @file h264.c
24
 * H.264 / AVC / MPEG4 part10 codec.
25
 * @author Michael Niedermayer <michaelni@gmx.at>
26
 */
27

    
28
#include "dsputil.h"
29
#include "avcodec.h"
30
#include "mpegvideo.h"
31
#include "h264.h"
32
#include "h264data.h"
33
#include "h264_parser.h"
34
#include "golomb.h"
35
#include "rectangle.h"
36

    
37
#include "cabac.h"
38
#ifdef ARCH_X86
39
#include "i386/h264_i386.h"
40
#endif
41

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

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

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

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

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

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

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

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

    
75
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
76
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
77
static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
78
static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
79
static Picture * remove_long(H264Context *h, int i, int ref_mask);
80

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

    
89
const uint8_t ff_rem6[52]={
90
0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
91
};
92

    
93
const uint8_t ff_div6[52]={
94
0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
95
};
96

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

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

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

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

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

    
122
    topleft_xy = top_xy - 1;
123
    topright_xy= top_xy + 1;
124
    left_xy[1] = left_xy[0] = mb_xy-1;
125
    left_block = left_block_options[0];
126
    if(FRAME_MBAFF){
127
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
128
        const int top_pair_xy      = pair_xy     - s->mb_stride;
129
        const int topleft_pair_xy  = top_pair_xy - 1;
130
        const int topright_pair_xy = top_pair_xy + 1;
131
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
132
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
133
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
134
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
135
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
136
        const int bottom = (s->mb_y & 1);
137
        tprintf(s->avctx, "fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
138
        if (bottom
139
                ? !curr_mb_frame_flag // bottom macroblock
140
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
141
                ) {
142
            top_xy -= s->mb_stride;
143
        }
144
        if (bottom
145
                ? !curr_mb_frame_flag // bottom macroblock
146
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
147
                ) {
148
            topleft_xy -= s->mb_stride;
149
        } else if(bottom && curr_mb_frame_flag && !left_mb_frame_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 (bottom
155
                ? !curr_mb_frame_flag // bottom macroblock
156
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
157
                ) {
158
            topright_xy -= s->mb_stride;
159
        }
160
        if (left_mb_frame_flag != curr_mb_frame_flag) {
161
            left_xy[1] = left_xy[0] = pair_xy - 1;
162
            if (curr_mb_frame_flag) {
163
                if (bottom) {
164
                    left_block = left_block_options[1];
165
                } else {
166
                    left_block= left_block_options[2];
167
                }
168
            } else {
169
                left_xy[1] += s->mb_stride;
170
                left_block = left_block_options[3];
171
            }
172
        }
173
    }
174

    
175
    h->top_mb_xy = top_xy;
176
    h->left_mb_xy[0] = left_xy[0];
177
    h->left_mb_xy[1] = left_xy[1];
178
    if(for_deblock){
179
        topleft_type = 0;
180
        topright_type = 0;
181
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
182
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
183
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
184

    
185
        if(MB_MBAFF && !IS_INTRA(mb_type)){
186
            int list;
187
            for(list=0; list<h->list_count; list++){
188
                //These values where changed for ease of performing MC, we need to change them back
189
                //FIXME maybe we can make MC and loop filter use the same values or prevent
190
                //the MC code from changing ref_cache and rather use a temporary array.
191
                if(USES_LIST(mb_type,list)){
192
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
193
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
194
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
195
                    ref += h->b8_stride;
196
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
197
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
198
                }
199
            }
200
        }
201
    }else{
202
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
203
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
204
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
205
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
206
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
207

    
208
    if(IS_INTRA(mb_type)){
209
        int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
210
        h->topleft_samples_available=
211
        h->top_samples_available=
212
        h->left_samples_available= 0xFFFF;
213
        h->topright_samples_available= 0xEEEA;
214

    
215
        if(!(top_type & type_mask)){
216
            h->topleft_samples_available= 0xB3FF;
217
            h->top_samples_available= 0x33FF;
218
            h->topright_samples_available= 0x26EA;
219
        }
220
        if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
221
            if(IS_INTERLACED(mb_type)){
222
                if(!(left_type[0] & type_mask)){
223
                    h->topleft_samples_available&= 0xDFFF;
224
                    h->left_samples_available&= 0x5FFF;
225
                }
226
                if(!(left_type[1] & type_mask)){
227
                    h->topleft_samples_available&= 0xFF5F;
228
                    h->left_samples_available&= 0xFF5F;
229
                }
230
            }else{
231
                int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
232
                                ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
233
                assert(left_xy[0] == left_xy[1]);
234
                if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
235
                    h->topleft_samples_available&= 0xDF5F;
236
                    h->left_samples_available&= 0x5F5F;
237
                }
238
            }
239
        }else{
240
            if(!(left_type[0] & type_mask)){
241
                h->topleft_samples_available&= 0xDF5F;
242
                h->left_samples_available&= 0x5F5F;
243
            }
244
        }
245

    
246
        if(!(topleft_type & type_mask))
247
            h->topleft_samples_available&= 0x7FFF;
248

    
249
        if(!(topright_type & type_mask))
250
            h->topright_samples_available&= 0xFBFF;
251

    
252
        if(IS_INTRA4x4(mb_type)){
253
            if(IS_INTRA4x4(top_type)){
254
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
255
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
256
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
257
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
258
            }else{
259
                int pred;
260
                if(!(top_type & type_mask))
261
                    pred= -1;
262
                else{
263
                    pred= 2;
264
                }
265
                h->intra4x4_pred_mode_cache[4+8*0]=
266
                h->intra4x4_pred_mode_cache[5+8*0]=
267
                h->intra4x4_pred_mode_cache[6+8*0]=
268
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
269
            }
270
            for(i=0; i<2; i++){
271
                if(IS_INTRA4x4(left_type[i])){
272
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
273
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
274
                }else{
275
                    int pred;
276
                    if(!(left_type[i] & type_mask))
277
                        pred= -1;
278
                    else{
279
                        pred= 2;
280
                    }
281
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
282
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
283
                }
284
            }
285
        }
286
    }
287
    }
288

    
289

    
290
/*
291
0 . T T. T T T T
292
1 L . .L . . . .
293
2 L . .L . . . .
294
3 . T TL . . . .
295
4 L . .L . . . .
296
5 L . .. . . . .
297
*/
298
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
299
    if(top_type){
300
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
301
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
302
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
303
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
304

    
305
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
306
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
307

    
308
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
309
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
310

    
311
    }else{
312
        h->non_zero_count_cache[4+8*0]=
313
        h->non_zero_count_cache[5+8*0]=
314
        h->non_zero_count_cache[6+8*0]=
315
        h->non_zero_count_cache[7+8*0]=
316

    
317
        h->non_zero_count_cache[1+8*0]=
318
        h->non_zero_count_cache[2+8*0]=
319

    
320
        h->non_zero_count_cache[1+8*3]=
321
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
322

    
323
    }
324

    
325
    for (i=0; i<2; i++) {
326
        if(left_type[i]){
327
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
328
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
329
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
330
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
331
        }else{
332
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
333
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
334
            h->non_zero_count_cache[0+8*1 +   8*i]=
335
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
336
        }
337
    }
338

    
339
    if( h->pps.cabac ) {
340
        // top_cbp
341
        if(top_type) {
342
            h->top_cbp = h->cbp_table[top_xy];
343
        } else if(IS_INTRA(mb_type)) {
344
            h->top_cbp = 0x1C0;
345
        } else {
346
            h->top_cbp = 0;
347
        }
348
        // left_cbp
349
        if (left_type[0]) {
350
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
351
        } else if(IS_INTRA(mb_type)) {
352
            h->left_cbp = 0x1C0;
353
        } else {
354
            h->left_cbp = 0;
355
        }
356
        if (left_type[0]) {
357
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
358
        }
359
        if (left_type[1]) {
360
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
361
        }
362
    }
363

    
364
#if 1
365
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
366
        int list;
367
        for(list=0; list<h->list_count; list++){
368
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
369
                /*if(!h->mv_cache_clean[list]){
370
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
371
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
372
                    h->mv_cache_clean[list]= 1;
373
                }*/
374
                continue;
375
            }
376
            h->mv_cache_clean[list]= 0;
377

    
378
            if(USES_LIST(top_type, list)){
379
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
380
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
381
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
382
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
383
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
384
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
385
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
386
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
387
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
388
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
389
            }else{
390
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
391
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
392
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
393
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
394
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
395
            }
396

    
397
            for(i=0; i<2; i++){
398
                int cache_idx = scan8[0] - 1 + i*2*8;
399
                if(USES_LIST(left_type[i], list)){
400
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
401
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
402
                    *(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]];
403
                    *(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]];
404
                    h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
405
                    h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
406
                }else{
407
                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
408
                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
409
                    h->ref_cache[list][cache_idx  ]=
410
                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
411
                }
412
            }
413

    
414
            if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
415
                continue;
416

    
417
            if(USES_LIST(topleft_type, list)){
418
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
419
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
420
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
421
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
422
            }else{
423
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
424
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
425
            }
426

    
427
            if(USES_LIST(topright_type, list)){
428
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
429
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
430
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
431
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
432
            }else{
433
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
434
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
435
            }
436

    
437
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
438
                continue;
439

    
440
            h->ref_cache[list][scan8[5 ]+1] =
441
            h->ref_cache[list][scan8[7 ]+1] =
442
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
443
            h->ref_cache[list][scan8[4 ]] =
444
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
445
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
446
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
447
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
448
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
449
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
450

    
451
            if( h->pps.cabac ) {
452
                /* XXX beurk, Load mvd */
453
                if(USES_LIST(top_type, list)){
454
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
455
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
456
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
457
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
458
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
459
                }else{
460
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
461
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
462
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
463
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
464
                }
465
                if(USES_LIST(left_type[0], list)){
466
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
467
                    *(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]];
468
                    *(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]];
469
                }else{
470
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
471
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
472
                }
473
                if(USES_LIST(left_type[1], list)){
474
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
475
                    *(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]];
476
                    *(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]];
477
                }else{
478
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
479
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
480
                }
481
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
482
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
483
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
484
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
485
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
486

    
487
                if(h->slice_type_nos == FF_B_TYPE){
488
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
489

    
490
                    if(IS_DIRECT(top_type)){
491
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
492
                    }else if(IS_8X8(top_type)){
493
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
494
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
495
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
496
                    }else{
497
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
498
                    }
499

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

    
507
                    if(IS_DIRECT(left_type[1]))
508
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
509
                    else if(IS_8X8(left_type[1]))
510
                        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)];
511
                    else
512
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
513
                }
514
            }
515

    
516
            if(FRAME_MBAFF){
517
#define MAP_MVS\
518
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
519
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
520
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
521
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
522
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
523
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
524
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
525
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
526
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
527
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
528
                if(MB_FIELD){
529
#define MAP_F2F(idx, mb_type)\
530
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
531
                        h->ref_cache[list][idx] <<= 1;\
532
                        h->mv_cache[list][idx][1] /= 2;\
533
                        h->mvd_cache[list][idx][1] /= 2;\
534
                    }
535
                    MAP_MVS
536
#undef MAP_F2F
537
                }else{
538
#define MAP_F2F(idx, mb_type)\
539
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
540
                        h->ref_cache[list][idx] >>= 1;\
541
                        h->mv_cache[list][idx][1] <<= 1;\
542
                        h->mvd_cache[list][idx][1] <<= 1;\
543
                    }
544
                    MAP_MVS
545
#undef MAP_F2F
546
                }
547
            }
548
        }
549
    }
550
#endif
551

    
552
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
553
}
554

    
555
static inline void write_back_intra_pred_mode(H264Context *h){
556
    const int mb_xy= h->mb_xy;
557

    
558
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
559
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
560
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
561
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
562
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
563
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
564
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
565
}
566

    
567
/**
568
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
569
 */
570
static inline int check_intra4x4_pred_mode(H264Context *h){
571
    MpegEncContext * const s = &h->s;
572
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
573
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
574
    int i;
575

    
576
    if(!(h->top_samples_available&0x8000)){
577
        for(i=0; i<4; i++){
578
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
579
            if(status<0){
580
                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);
581
                return -1;
582
            } else if(status){
583
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
584
            }
585
        }
586
    }
587

    
588
    if((h->left_samples_available&0x8888)!=0x8888){
589
        static const int mask[4]={0x8000,0x2000,0x80,0x20};
590
        for(i=0; i<4; i++){
591
            if(!(h->left_samples_available&mask[i])){
592
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
593
            if(status<0){
594
                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);
595
                return -1;
596
            } else if(status){
597
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
598
            }
599
            }
600
        }
601
    }
602

    
603
    return 0;
604
} //FIXME cleanup like next
605

    
606
/**
607
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
608
 */
609
static inline int check_intra_pred_mode(H264Context *h, int mode){
610
    MpegEncContext * const s = &h->s;
611
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
612
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
613

    
614
    if(mode > 6U) {
615
        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);
616
        return -1;
617
    }
618

    
619
    if(!(h->top_samples_available&0x8000)){
620
        mode= top[ mode ];
621
        if(mode<0){
622
            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);
623
            return -1;
624
        }
625
    }
626

    
627
    if((h->left_samples_available&0x8080) != 0x8080){
628
        mode= left[ mode ];
629
        if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
630
            mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
631
        }
632
        if(mode<0){
633
            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);
634
            return -1;
635
        }
636
    }
637

    
638
    return mode;
639
}
640

    
641
/**
642
 * gets the predicted intra4x4 prediction mode.
643
 */
644
static inline int pred_intra_mode(H264Context *h, int n){
645
    const int index8= scan8[n];
646
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
647
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
648
    const int min= FFMIN(left, top);
649

    
650
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
651

    
652
    if(min<0) return DC_PRED;
653
    else      return min;
654
}
655

    
656
static inline void write_back_non_zero_count(H264Context *h){
657
    const int mb_xy= h->mb_xy;
658

    
659
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
660
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
661
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
662
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
663
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
664
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
665
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
666

    
667
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
668
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
669
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
670

    
671
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
672
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
673
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
674
}
675

    
676
/**
677
 * gets the predicted number of non-zero coefficients.
678
 * @param n block index
679
 */
680
static inline int pred_non_zero_count(H264Context *h, int n){
681
    const int index8= scan8[n];
682
    const int left= h->non_zero_count_cache[index8 - 1];
683
    const int top = h->non_zero_count_cache[index8 - 8];
684
    int i= left + top;
685

    
686
    if(i<64) i= (i+1)>>1;
687

    
688
    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
689

    
690
    return i&31;
691
}
692

    
693
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
694
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
695
    MpegEncContext *s = &h->s;
696

    
697
    /* there is no consistent mapping of mvs to neighboring locations that will
698
     * make mbaff happy, so we can't move all this logic to fill_caches */
699
    if(FRAME_MBAFF){
700
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
701
        const int16_t *mv;
702
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
703
        *C = h->mv_cache[list][scan8[0]-2];
704

    
705
        if(!MB_FIELD
706
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
707
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
708
            if(IS_INTERLACED(mb_types[topright_xy])){
709
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
710
                const int x4 = X4, y4 = Y4;\
711
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
712
                if(!USES_LIST(mb_type,list))\
713
                    return LIST_NOT_USED;\
714
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
715
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
716
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
717
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
718

    
719
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
720
            }
721
        }
722
        if(topright_ref == PART_NOT_AVAILABLE
723
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
724
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
725
            if(!MB_FIELD
726
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
727
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
728
            }
729
            if(MB_FIELD
730
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
731
               && i >= scan8[0]+8){
732
                // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
733
                SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
734
            }
735
        }
736
#undef SET_DIAG_MV
737
    }
738

    
739
    if(topright_ref != PART_NOT_AVAILABLE){
740
        *C= h->mv_cache[list][ i - 8 + part_width ];
741
        return topright_ref;
742
    }else{
743
        tprintf(s->avctx, "topright MV not available\n");
744

    
745
        *C= h->mv_cache[list][ i - 8 - 1 ];
746
        return h->ref_cache[list][ i - 8 - 1 ];
747
    }
748
}
749

    
750
/**
751
 * gets the predicted MV.
752
 * @param n the block index
753
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
754
 * @param mx the x component of the predicted motion vector
755
 * @param my the y component of the predicted motion vector
756
 */
757
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
758
    const int index8= scan8[n];
759
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
760
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
761
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
762
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
763
    const int16_t * C;
764
    int diagonal_ref, match_count;
765

    
766
    assert(part_width==1 || part_width==2 || part_width==4);
767

    
768
/* mv_cache
769
  B . . A T T T T
770
  U . . L . . , .
771
  U . . L . . . .
772
  U . . L . . , .
773
  . . . L . . . .
774
*/
775

    
776
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
777
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
778
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
779
    if(match_count > 1){ //most common
780
        *mx= mid_pred(A[0], B[0], C[0]);
781
        *my= mid_pred(A[1], B[1], C[1]);
782
    }else if(match_count==1){
783
        if(left_ref==ref){
784
            *mx= A[0];
785
            *my= A[1];
786
        }else if(top_ref==ref){
787
            *mx= B[0];
788
            *my= B[1];
789
        }else{
790
            *mx= C[0];
791
            *my= C[1];
792
        }
793
    }else{
794
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
795
            *mx= A[0];
796
            *my= A[1];
797
        }else{
798
            *mx= mid_pred(A[0], B[0], C[0]);
799
            *my= mid_pred(A[1], B[1], C[1]);
800
        }
801
    }
802

    
803
    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);
804
}
805

    
806
/**
807
 * gets the directionally predicted 16x8 MV.
808
 * @param n the block index
809
 * @param mx the x component of the predicted motion vector
810
 * @param my the y component of the predicted motion vector
811
 */
812
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
813
    if(n==0){
814
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
815
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
816

    
817
        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);
818

    
819
        if(top_ref == ref){
820
            *mx= B[0];
821
            *my= B[1];
822
            return;
823
        }
824
    }else{
825
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
826
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
827

    
828
        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);
829

    
830
        if(left_ref == ref){
831
            *mx= A[0];
832
            *my= A[1];
833
            return;
834
        }
835
    }
836

    
837
    //RARE
838
    pred_motion(h, n, 4, list, ref, mx, my);
839
}
840

    
841
/**
842
 * gets the directionally predicted 8x16 MV.
843
 * @param n the block index
844
 * @param mx the x component of the predicted motion vector
845
 * @param my the y component of the predicted motion vector
846
 */
847
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
848
    if(n==0){
849
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
850
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
851

    
852
        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);
853

    
854
        if(left_ref == ref){
855
            *mx= A[0];
856
            *my= A[1];
857
            return;
858
        }
859
    }else{
860
        const int16_t * C;
861
        int diagonal_ref;
862

    
863
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
864

    
865
        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);
866

    
867
        if(diagonal_ref == ref){
868
            *mx= C[0];
869
            *my= C[1];
870
            return;
871
        }
872
    }
873

    
874
    //RARE
875
    pred_motion(h, n, 2, list, ref, mx, my);
876
}
877

    
878
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
879
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
880
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
881

    
882
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
883

    
884
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
885
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
886
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
887

    
888
        *mx = *my = 0;
889
        return;
890
    }
891

    
892
    pred_motion(h, 0, 4, 0, 0, mx, my);
893

    
894
    return;
895
}
896

    
897
static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
898
    int poc0 = h->ref_list[0][i].poc;
899
    int td = av_clip(poc1 - poc0, -128, 127);
900
    if(td == 0 || h->ref_list[0][i].long_ref){
901
        return 256;
902
    }else{
903
        int tb = av_clip(poc - poc0, -128, 127);
904
        int tx = (16384 + (FFABS(td) >> 1)) / td;
905
        return av_clip((tb*tx + 32) >> 6, -1024, 1023);
906
    }
907
}
908

    
909
static inline void direct_dist_scale_factor(H264Context * const h){
910
    MpegEncContext * const s = &h->s;
911
    const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
912
    const int poc1 = h->ref_list[1][0].poc;
913
    int i, field;
914
    for(field=0; field<2; field++){
915
        const int poc  = h->s.current_picture_ptr->field_poc[field];
916
        const int poc1 = h->ref_list[1][0].field_poc[field];
917
        for(i=0; i < 2*h->ref_count[0]; i++)
918
            h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
919
    }
920

    
921
    for(i=0; i<h->ref_count[0]; i++){
922
        h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
923
    }
924
}
925
static inline void direct_ref_list_init(H264Context * const h){
926
    MpegEncContext * const s = &h->s;
927
    Picture * const ref1 = &h->ref_list[1][0];
928
    Picture * const cur = s->current_picture_ptr;
929
    int list, i, j, field, rfield;
930
    int sidx= s->picture_structure&1;
931
    int ref1sidx= ref1->reference&1;
932
    for(list=0; list<2; list++){
933
        cur->ref_count[sidx][list] = h->ref_count[list];
934
        for(j=0; j<h->ref_count[list]; j++)
935
            cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
936
    }
937
    if(s->picture_structure == PICT_FRAME){
938
        memcpy(cur->ref_count[0], cur->ref_count[1], sizeof(cur->ref_count[0]));
939
        memcpy(cur->ref_poc  [0], cur->ref_poc  [1], sizeof(cur->ref_poc  [0]));
940
    }
941
    if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
942
        return;
943
    for(list=0; list<2; list++){
944
        for(field=0; field<2; field++){
945
            for(i=0; i<ref1->ref_count[field][list]; i++){
946
                for(rfield=0; rfield<2; rfield++){
947
                    int poc = ref1->ref_poc[field][list][i];
948
                    if((poc&3) == 3)
949
                        poc= (poc&~3) + rfield + 1;
950

    
951
                    h->map_col_to_list0_field[field][list][2*i+rfield] = 0; /* bogus; fills in for missing frames */
952
                    for(j=16; j<16+2*h->ref_count[list]; j++)
953
                        if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
954
                            h->map_col_to_list0_field[field][list][2*i+rfield] = j-16;
955
                            break;
956
                        }
957
                }
958
            }
959
        }
960

    
961
        for(i=0; i<ref1->ref_count[ref1sidx][list]; i++){
962
            int poc = ref1->ref_poc[ref1sidx][list][i];
963
            if(((poc&3) == 3) != (s->picture_structure == PICT_FRAME))
964
                poc= (poc&~3) + s->picture_structure;
965
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
966
            for(j=0; j<h->ref_count[list]; j++)
967
                if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
968
                    h->map_col_to_list0[list][i] = j;
969
                    break;
970
                }
971
        }
972
    }
973
}
974

    
975
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
976
    MpegEncContext * const s = &h->s;
977
    int b8_stride = h->b8_stride;
978
    int b4_stride = h->b_stride;
979
    int mb_xy = h->mb_xy;
980
    int mb_type_col[2];
981
    const int16_t (*l1mv0)[2], (*l1mv1)[2];
982
    const int8_t *l1ref0, *l1ref1;
983
    const int is_b8x8 = IS_8X8(*mb_type);
984
    unsigned int sub_mb_type;
985
    int i8, i4;
986

    
987
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
988

    
989
    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
990
        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
991
                int cur_poc = s->current_picture_ptr->poc;
992
                int *col_poc = h->ref_list[1]->field_poc;
993
                int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
994
                mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
995
                b8_stride = 0;
996
        }else if(!(s->picture_structure & h->ref_list[1][0].reference)){// FL -> FL & differ parity
997
            int fieldoff= 2*(h->ref_list[1][0].reference)-3;
998
            mb_xy += s->mb_stride*fieldoff;
999
        }
1000
        goto single_col;
1001
    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
1002
        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
1003
            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
1004
            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
1005
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
1006
            b8_stride *= 3;
1007
            b4_stride *= 6;
1008
            //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag
1009
            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
1010
                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
1011
                && !is_b8x8){
1012
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1013
                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
1014
            }else{
1015
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1016
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1017
            }
1018
        }else{                                           //     AFR/FR    -> AFR/FR
1019
single_col:
1020
            mb_type_col[0] =
1021
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
1022
            if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
1023
                /* FIXME save sub mb types from previous frames (or derive from MVs)
1024
                * so we know exactly what block size to use */
1025
                sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1026
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1027
            }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
1028
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1029
                *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1030
            }else{
1031
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1032
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1033
            }
1034
        }
1035
    }
1036

    
1037
    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
1038
    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
1039
    l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
1040
    l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
1041
    if(!b8_stride){
1042
        if(s->mb_y&1){
1043
            l1ref0 += h->b8_stride;
1044
            l1ref1 += h->b8_stride;
1045
            l1mv0  +=  2*b4_stride;
1046
            l1mv1  +=  2*b4_stride;
1047
        }
1048
    }
1049

    
1050
    if(h->direct_spatial_mv_pred){
1051
        int ref[2];
1052
        int mv[2][2];
1053
        int list;
1054

    
1055
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1056

    
1057
        /* ref = min(neighbors) */
1058
        for(list=0; list<2; list++){
1059
            int refa = h->ref_cache[list][scan8[0] - 1];
1060
            int refb = h->ref_cache[list][scan8[0] - 8];
1061
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1062
            if(refc == PART_NOT_AVAILABLE)
1063
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1064
            ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
1065
            if(ref[list] < 0)
1066
                ref[list] = -1;
1067
        }
1068

    
1069
        if(ref[0] < 0 && ref[1] < 0){
1070
            ref[0] = ref[1] = 0;
1071
            mv[0][0] = mv[0][1] =
1072
            mv[1][0] = mv[1][1] = 0;
1073
        }else{
1074
            for(list=0; list<2; list++){
1075
                if(ref[list] >= 0)
1076
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1077
                else
1078
                    mv[list][0] = mv[list][1] = 0;
1079
            }
1080
        }
1081

    
1082
        if(ref[1] < 0){
1083
            if(!is_b8x8)
1084
                *mb_type &= ~MB_TYPE_L1;
1085
            sub_mb_type &= ~MB_TYPE_L1;
1086
        }else if(ref[0] < 0){
1087
            if(!is_b8x8)
1088
                *mb_type &= ~MB_TYPE_L0;
1089
            sub_mb_type &= ~MB_TYPE_L0;
1090
        }
1091

    
1092
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1093
            for(i8=0; i8<4; i8++){
1094
                int x8 = i8&1;
1095
                int y8 = i8>>1;
1096
                int xy8 = x8+y8*b8_stride;
1097
                int xy4 = 3*x8+y8*b4_stride;
1098
                int a=0, b=0;
1099

    
1100
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1101
                    continue;
1102
                h->sub_mb_type[i8] = sub_mb_type;
1103

    
1104
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1105
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1106
                if(!IS_INTRA(mb_type_col[y8])
1107
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1108
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1109
                    if(ref[0] > 0)
1110
                        a= pack16to32(mv[0][0],mv[0][1]);
1111
                    if(ref[1] > 0)
1112
                        b= pack16to32(mv[1][0],mv[1][1]);
1113
                }else{
1114
                    a= pack16to32(mv[0][0],mv[0][1]);
1115
                    b= pack16to32(mv[1][0],mv[1][1]);
1116
                }
1117
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1118
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1119
            }
1120
        }else if(IS_16X16(*mb_type)){
1121
            int a=0, b=0;
1122

    
1123
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1124
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1125
            if(!IS_INTRA(mb_type_col[0])
1126
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1127
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1128
                       && (h->x264_build>33 || !h->x264_build)))){
1129
                if(ref[0] > 0)
1130
                    a= pack16to32(mv[0][0],mv[0][1]);
1131
                if(ref[1] > 0)
1132
                    b= pack16to32(mv[1][0],mv[1][1]);
1133
            }else{
1134
                a= pack16to32(mv[0][0],mv[0][1]);
1135
                b= pack16to32(mv[1][0],mv[1][1]);
1136
            }
1137
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1138
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1139
        }else{
1140
            for(i8=0; i8<4; i8++){
1141
                const int x8 = i8&1;
1142
                const int y8 = i8>>1;
1143

    
1144
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1145
                    continue;
1146
                h->sub_mb_type[i8] = sub_mb_type;
1147

    
1148
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1149
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1150
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1151
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1152

    
1153
                /* col_zero_flag */
1154
                if(!IS_INTRA(mb_type_col[0]) && (   l1ref0[x8 + y8*b8_stride] == 0
1155
                                              || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
1156
                                                  && (h->x264_build>33 || !h->x264_build)))){
1157
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
1158
                    if(IS_SUB_8X8(sub_mb_type)){
1159
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1160
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1161
                            if(ref[0] == 0)
1162
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1163
                            if(ref[1] == 0)
1164
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1165
                        }
1166
                    }else
1167
                    for(i4=0; i4<4; i4++){
1168
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1169
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1170
                            if(ref[0] == 0)
1171
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1172
                            if(ref[1] == 0)
1173
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1174
                        }
1175
                    }
1176
                }
1177
            }
1178
        }
1179
    }else{ /* direct temporal mv pred */
1180
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1181
        const int *dist_scale_factor = h->dist_scale_factor;
1182

    
1183
        if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
1184
            map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
1185
            map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
1186
            dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];
1187
        }
1188
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1189
            /* FIXME assumes direct_8x8_inference == 1 */
1190
            int y_shift  = 2*!IS_INTERLACED(*mb_type);
1191
            int ref_shift= FRAME_MBAFF ? y_shift : 1;
1192

    
1193
            for(i8=0; i8<4; i8++){
1194
                const int x8 = i8&1;
1195
                const int y8 = i8>>1;
1196
                int ref0, scale;
1197
                const int16_t (*l1mv)[2]= l1mv0;
1198

    
1199
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1200
                    continue;
1201
                h->sub_mb_type[i8] = sub_mb_type;
1202

    
1203
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1204
                if(IS_INTRA(mb_type_col[y8])){
1205
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1206
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1207
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1208
                    continue;
1209
                }
1210

    
1211
                ref0 = l1ref0[x8 + y8*b8_stride];
1212
                if(ref0 >= 0)
1213
                    ref0 = map_col_to_list0[0][ref0*2>>ref_shift];
1214
                else{
1215
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride]*2>>ref_shift];
1216
                    l1mv= l1mv1;
1217
                }
1218
                scale = dist_scale_factor[ref0];
1219
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1220

    
1221
                {
1222
                    const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
1223
                    int my_col = (mv_col[1]<<y_shift)/2;
1224
                    int mx = (scale * mv_col[0] + 128) >> 8;
1225
                    int my = (scale * my_col + 128) >> 8;
1226
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1227
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1228
                }
1229
            }
1230
            return;
1231
        }
1232

    
1233
        /* one-to-one mv scaling */
1234

    
1235
        if(IS_16X16(*mb_type)){
1236
            int ref, mv0, mv1;
1237

    
1238
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1239
            if(IS_INTRA(mb_type_col[0])){
1240
                ref=mv0=mv1=0;
1241
            }else{
1242
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1243
                                                : map_col_to_list0[1][l1ref1[0]];
1244
                const int scale = dist_scale_factor[ref0];
1245
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1246
                int mv_l0[2];
1247
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1248
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1249
                ref= ref0;
1250
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
1251
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1252
            }
1253
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1254
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1255
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1256
        }else{
1257
            for(i8=0; i8<4; i8++){
1258
                const int x8 = i8&1;
1259
                const int y8 = i8>>1;
1260
                int ref0, scale;
1261
                const int16_t (*l1mv)[2]= l1mv0;
1262

    
1263
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1264
                    continue;
1265
                h->sub_mb_type[i8] = sub_mb_type;
1266
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1267
                if(IS_INTRA(mb_type_col[0])){
1268
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1269
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1270
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1271
                    continue;
1272
                }
1273

    
1274
                ref0 = l1ref0[x8 + y8*b8_stride];
1275
                if(ref0 >= 0)
1276
                    ref0 = map_col_to_list0[0][ref0];
1277
                else{
1278
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride]];
1279
                    l1mv= l1mv1;
1280
                }
1281
                scale = dist_scale_factor[ref0];
1282

    
1283
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1284
                if(IS_SUB_8X8(sub_mb_type)){
1285
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1286
                    int mx = (scale * mv_col[0] + 128) >> 8;
1287
                    int my = (scale * mv_col[1] + 128) >> 8;
1288
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1289
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1290
                }else
1291
                for(i4=0; i4<4; i4++){
1292
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1293
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1294
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1295
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1296
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1297
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1298
                }
1299
            }
1300
        }
1301
    }
1302
}
1303

    
1304
static inline void write_back_motion(H264Context *h, int mb_type){
1305
    MpegEncContext * const s = &h->s;
1306
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1307
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1308
    int list;
1309

    
1310
    if(!USES_LIST(mb_type, 0))
1311
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1312

    
1313
    for(list=0; list<h->list_count; list++){
1314
        int y;
1315
        if(!USES_LIST(mb_type, list))
1316
            continue;
1317

    
1318
        for(y=0; y<4; y++){
1319
            *(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];
1320
            *(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];
1321
        }
1322
        if( h->pps.cabac ) {
1323
            if(IS_SKIP(mb_type))
1324
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1325
            else
1326
            for(y=0; y<4; y++){
1327
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1328
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1329
            }
1330
        }
1331

    
1332
        {
1333
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1334
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1335
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1336
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1337
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1338
        }
1339
    }
1340

    
1341
    if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
1342
        if(IS_8X8(mb_type)){
1343
            uint8_t *direct_table = &h->direct_table[b8_xy];
1344
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1345
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1346
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1347
        }
1348
    }
1349
}
1350

    
1351
/**
1352
 * Decodes a network abstraction layer unit.
1353
 * @param consumed is the number of bytes used as input
1354
 * @param length is the length of the array
1355
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1356
 * @returns decoded bytes, might be src+1 if no escapes
1357
 */
1358
static const uint8_t *decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1359
    int i, si, di;
1360
    uint8_t *dst;
1361
    int bufidx;
1362

    
1363
//    src[0]&0x80;                //forbidden bit
1364
    h->nal_ref_idc= src[0]>>5;
1365
    h->nal_unit_type= src[0]&0x1F;
1366

    
1367
    src++; length--;
1368
#if 0
1369
    for(i=0; i<length; i++)
1370
        printf("%2X ", src[i]);
1371
#endif
1372
    for(i=0; i+1<length; i+=2){
1373
        if(src[i]) continue;
1374
        if(i>0 && src[i-1]==0) i--;
1375
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1376
            if(src[i+2]!=3){
1377
                /* startcode, so we must be past the end */
1378
                length=i;
1379
            }
1380
            break;
1381
        }
1382
    }
1383

    
1384
    if(i>=length-1){ //no escaped 0
1385
        *dst_length= length;
1386
        *consumed= length+1; //+1 for the header
1387
        return src;
1388
    }
1389

    
1390
    bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1391
    h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length);
1392
    dst= h->rbsp_buffer[bufidx];
1393

    
1394
    if (dst == NULL){
1395
        return NULL;
1396
    }
1397

    
1398
//printf("decoding esc\n");
1399
    si=di=0;
1400
    while(si<length){
1401
        //remove escapes (very rare 1:2^22)
1402
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1403
            if(src[si+2]==3){ //escape
1404
                dst[di++]= 0;
1405
                dst[di++]= 0;
1406
                si+=3;
1407
                continue;
1408
            }else //next start code
1409
                break;
1410
        }
1411

    
1412
        dst[di++]= src[si++];
1413
    }
1414

    
1415
    *dst_length= di;
1416
    *consumed= si + 1;//+1 for the header
1417
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1418
    return dst;
1419
}
1420

    
1421
/**
1422
 * identifies the exact end of the bitstream
1423
 * @return the length of the trailing, or 0 if damaged
1424
 */
1425
static int decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1426
    int v= *src;
1427
    int r;
1428

    
1429
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1430

    
1431
    for(r=1; r<9; r++){
1432
        if(v&1) return r;
1433
        v>>=1;
1434
    }
1435
    return 0;
1436
}
1437

    
1438
/**
1439
 * IDCT transforms the 16 dc values and dequantizes them.
1440
 * @param qp quantization parameter
1441
 */
1442
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1443
#define stride 16
1444
    int i;
1445
    int temp[16]; //FIXME check if this is a good idea
1446
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1447
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1448

    
1449
//memset(block, 64, 2*256);
1450
//return;
1451
    for(i=0; i<4; i++){
1452
        const int offset= y_offset[i];
1453
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1454
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1455
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1456
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1457

    
1458
        temp[4*i+0]= z0+z3;
1459
        temp[4*i+1]= z1+z2;
1460
        temp[4*i+2]= z1-z2;
1461
        temp[4*i+3]= z0-z3;
1462
    }
1463

    
1464
    for(i=0; i<4; i++){
1465
        const int offset= x_offset[i];
1466
        const int z0= temp[4*0+i] + temp[4*2+i];
1467
        const int z1= temp[4*0+i] - temp[4*2+i];
1468
        const int z2= temp[4*1+i] - temp[4*3+i];
1469
        const int z3= temp[4*1+i] + temp[4*3+i];
1470

    
1471
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
1472
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1473
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1474
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1475
    }
1476
}
1477

    
1478
#if 0
1479
/**
1480
 * DCT transforms the 16 dc values.
1481
 * @param qp quantization parameter ??? FIXME
1482
 */
1483
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1484
//    const int qmul= dequant_coeff[qp][0];
1485
    int i;
1486
    int temp[16]; //FIXME check if this is a good idea
1487
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1488
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1489

1490
    for(i=0; i<4; i++){
1491
        const int offset= y_offset[i];
1492
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1493
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1494
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1495
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1496

1497
        temp[4*i+0]= z0+z3;
1498
        temp[4*i+1]= z1+z2;
1499
        temp[4*i+2]= z1-z2;
1500
        temp[4*i+3]= z0-z3;
1501
    }
1502

1503
    for(i=0; i<4; i++){
1504
        const int offset= x_offset[i];
1505
        const int z0= temp[4*0+i] + temp[4*2+i];
1506
        const int z1= temp[4*0+i] - temp[4*2+i];
1507
        const int z2= temp[4*1+i] - temp[4*3+i];
1508
        const int z3= temp[4*1+i] + temp[4*3+i];
1509

1510
        block[stride*0 +offset]= (z0 + z3)>>1;
1511
        block[stride*2 +offset]= (z1 + z2)>>1;
1512
        block[stride*8 +offset]= (z1 - z2)>>1;
1513
        block[stride*10+offset]= (z0 - z3)>>1;
1514
    }
1515
}
1516
#endif
1517

    
1518
#undef xStride
1519
#undef stride
1520

    
1521
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1522
    const int stride= 16*2;
1523
    const int xStride= 16;
1524
    int a,b,c,d,e;
1525

    
1526
    a= block[stride*0 + xStride*0];
1527
    b= block[stride*0 + xStride*1];
1528
    c= block[stride*1 + xStride*0];
1529
    d= block[stride*1 + xStride*1];
1530

    
1531
    e= a-b;
1532
    a= a+b;
1533
    b= c-d;
1534
    c= c+d;
1535

    
1536
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1537
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1538
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1539
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1540
}
1541

    
1542
#if 0
1543
static void chroma_dc_dct_c(DCTELEM *block){
1544
    const int stride= 16*2;
1545
    const int xStride= 16;
1546
    int a,b,c,d,e;
1547

1548
    a= block[stride*0 + xStride*0];
1549
    b= block[stride*0 + xStride*1];
1550
    c= block[stride*1 + xStride*0];
1551
    d= block[stride*1 + xStride*1];
1552

1553
    e= a-b;
1554
    a= a+b;
1555
    b= c-d;
1556
    c= c+d;
1557

1558
    block[stride*0 + xStride*0]= (a+c);
1559
    block[stride*0 + xStride*1]= (e+b);
1560
    block[stride*1 + xStride*0]= (a-c);
1561
    block[stride*1 + xStride*1]= (e-b);
1562
}
1563
#endif
1564

    
1565
/**
1566
 * gets the chroma qp.
1567
 */
1568
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1569
    return h->pps.chroma_qp_table[t][qscale];
1570
}
1571

    
1572
//FIXME need to check that this does not overflow signed 32 bit for low qp, I am not sure, it's very close
1573
//FIXME check that gcc inlines this (and optimizes intra & separate_dc stuff away)
1574
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int separate_dc){
1575
    int i;
1576
    const int * const quant_table= quant_coeff[qscale];
1577
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1578
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1579
    const unsigned int threshold2= (threshold1<<1);
1580
    int last_non_zero;
1581

    
1582
    if(separate_dc){
1583
        if(qscale<=18){
1584
            //avoid overflows
1585
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1586
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1587
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1588

    
1589
            int level= block[0]*quant_coeff[qscale+18][0];
1590
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1591
                if(level>0){
1592
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1593
                    block[0]= level;
1594
                }else{
1595
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1596
                    block[0]= -level;
1597
                }
1598
//                last_non_zero = i;
1599
            }else{
1600
                block[0]=0;
1601
            }
1602
        }else{
1603
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1604
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1605
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1606

    
1607
            int level= block[0]*quant_table[0];
1608
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1609
                if(level>0){
1610
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1611
                    block[0]= level;
1612
                }else{
1613
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1614
                    block[0]= -level;
1615
                }
1616
//                last_non_zero = i;
1617
            }else{
1618
                block[0]=0;
1619
            }
1620
        }
1621
        last_non_zero= 0;
1622
        i=1;
1623
    }else{
1624
        last_non_zero= -1;
1625
        i=0;
1626
    }
1627

    
1628
    for(; i<16; i++){
1629
        const int j= scantable[i];
1630
        int level= block[j]*quant_table[j];
1631

    
1632
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1633
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1634
        if(((unsigned)(level+threshold1))>threshold2){
1635
            if(level>0){
1636
                level= (bias + level)>>QUANT_SHIFT;
1637
                block[j]= level;
1638
            }else{
1639
                level= (bias - level)>>QUANT_SHIFT;
1640
                block[j]= -level;
1641
            }
1642
            last_non_zero = i;
1643
        }else{
1644
            block[j]=0;
1645
        }
1646
    }
1647

    
1648
    return last_non_zero;
1649
}
1650

    
1651
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1652
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1653
                           int src_x_offset, int src_y_offset,
1654
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1655
    MpegEncContext * const s = &h->s;
1656
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1657
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1658
    const int luma_xy= (mx&3) + ((my&3)<<2);
1659
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1660
    uint8_t * src_cb, * src_cr;
1661
    int extra_width= h->emu_edge_width;
1662
    int extra_height= h->emu_edge_height;
1663
    int emu=0;
1664
    const int full_mx= mx>>2;
1665
    const int full_my= my>>2;
1666
    const int pic_width  = 16*s->mb_width;
1667
    const int pic_height = 16*s->mb_height >> MB_FIELD;
1668

    
1669
    if(!pic->data[0]) //FIXME this is unacceptable, some sensible error concealment must be done for missing reference frames
1670
        return;
1671

    
1672
    if(mx&7) extra_width -= 3;
1673
    if(my&7) extra_height -= 3;
1674

    
1675
    if(   full_mx < 0-extra_width
1676
       || full_my < 0-extra_height
1677
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
1678
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
1679
        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);
1680
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1681
        emu=1;
1682
    }
1683

    
1684
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1685
    if(!square){
1686
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1687
    }
1688

    
1689
    if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1690

    
1691
    if(MB_FIELD){
1692
        // chroma offset when predicting from a field of opposite parity
1693
        my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1694
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1695
    }
1696
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1697
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1698

    
1699
    if(emu){
1700
        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);
1701
            src_cb= s->edge_emu_buffer;
1702
    }
1703
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1704

    
1705
    if(emu){
1706
        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);
1707
            src_cr= s->edge_emu_buffer;
1708
    }
1709
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1710
}
1711

    
1712
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1713
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1714
                           int x_offset, int y_offset,
1715
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1716
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1717
                           int list0, int list1){
1718
    MpegEncContext * const s = &h->s;
1719
    qpel_mc_func *qpix_op=  qpix_put;
1720
    h264_chroma_mc_func chroma_op= chroma_put;
1721

    
1722
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1723
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1724
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1725
    x_offset += 8*s->mb_x;
1726
    y_offset += 8*(s->mb_y >> MB_FIELD);
1727

    
1728
    if(list0){
1729
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1730
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1731
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1732
                           qpix_op, chroma_op);
1733

    
1734
        qpix_op=  qpix_avg;
1735
        chroma_op= chroma_avg;
1736
    }
1737

    
1738
    if(list1){
1739
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1740
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1741
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1742
                           qpix_op, chroma_op);
1743
    }
1744
}
1745

    
1746
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1747
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1748
                           int x_offset, int y_offset,
1749
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1750
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1751
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1752
                           int list0, int list1){
1753
    MpegEncContext * const s = &h->s;
1754

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

    
1761
    if(list0 && list1){
1762
        /* don't optimize for luma-only case, since B-frames usually
1763
         * use implicit weights => chroma too. */
1764
        uint8_t *tmp_cb = s->obmc_scratchpad;
1765
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1766
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1767
        int refn0 = h->ref_cache[0][ scan8[n] ];
1768
        int refn1 = h->ref_cache[1][ scan8[n] ];
1769

    
1770
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1771
                    dest_y, dest_cb, dest_cr,
1772
                    x_offset, y_offset, qpix_put, chroma_put);
1773
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1774
                    tmp_y, tmp_cb, tmp_cr,
1775
                    x_offset, y_offset, qpix_put, chroma_put);
1776

    
1777
        if(h->use_weight == 2){
1778
            int weight0 = h->implicit_weight[refn0][refn1];
1779
            int weight1 = 64 - weight0;
1780
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
1781
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1782
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1783
        }else{
1784
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1785
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1786
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1787
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1788
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1789
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1790
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1791
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1792
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1793
        }
1794
    }else{
1795
        int list = list1 ? 1 : 0;
1796
        int refn = h->ref_cache[list][ scan8[n] ];
1797
        Picture *ref= &h->ref_list[list][refn];
1798
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1799
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
1800
                    qpix_put, chroma_put);
1801

    
1802
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1803
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
1804
        if(h->use_weight_chroma){
1805
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1806
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1807
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1808
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1809
        }
1810
    }
1811
}
1812

    
1813
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1814
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1815
                           int x_offset, int y_offset,
1816
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1817
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1818
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1819
                           int list0, int list1){
1820
    if((h->use_weight==2 && list0 && list1
1821
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1822
       || h->use_weight==1)
1823
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1824
                         x_offset, y_offset, qpix_put, chroma_put,
1825
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1826
    else
1827
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1828
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1829
}
1830

    
1831
static inline void prefetch_motion(H264Context *h, int list){
1832
    /* fetch pixels for estimated mv 4 macroblocks ahead
1833
     * optimized for 64byte cache lines */
1834
    MpegEncContext * const s = &h->s;
1835
    const int refn = h->ref_cache[list][scan8[0]];
1836
    if(refn >= 0){
1837
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1838
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1839
        uint8_t **src= h->ref_list[list][refn].data;
1840
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1841
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
1842
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1843
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1844
    }
1845
}
1846

    
1847
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1848
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1849
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1850
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1851
    MpegEncContext * const s = &h->s;
1852
    const int mb_xy= h->mb_xy;
1853
    const int mb_type= s->current_picture.mb_type[mb_xy];
1854

    
1855
    assert(IS_INTER(mb_type));
1856

    
1857
    prefetch_motion(h, 0);
1858

    
1859
    if(IS_16X16(mb_type)){
1860
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1861
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1862
                &weight_op[0], &weight_avg[0],
1863
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1864
    }else if(IS_16X8(mb_type)){
1865
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1866
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1867
                &weight_op[1], &weight_avg[1],
1868
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1869
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1870
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1871
                &weight_op[1], &weight_avg[1],
1872
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1873
    }else if(IS_8X16(mb_type)){
1874
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1875
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1876
                &weight_op[2], &weight_avg[2],
1877
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1878
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1879
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1880
                &weight_op[2], &weight_avg[2],
1881
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1882
    }else{
1883
        int i;
1884

    
1885
        assert(IS_8X8(mb_type));
1886

    
1887
        for(i=0; i<4; i++){
1888
            const int sub_mb_type= h->sub_mb_type[i];
1889
            const int n= 4*i;
1890
            int x_offset= (i&1)<<2;
1891
            int y_offset= (i&2)<<1;
1892

    
1893
            if(IS_SUB_8X8(sub_mb_type)){
1894
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1895
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1896
                    &weight_op[3], &weight_avg[3],
1897
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1898
            }else if(IS_SUB_8X4(sub_mb_type)){
1899
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1900
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1901
                    &weight_op[4], &weight_avg[4],
1902
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1903
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1904
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1905
                    &weight_op[4], &weight_avg[4],
1906
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1907
            }else if(IS_SUB_4X8(sub_mb_type)){
1908
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1909
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1910
                    &weight_op[5], &weight_avg[5],
1911
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1912
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1913
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1914
                    &weight_op[5], &weight_avg[5],
1915
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1916
            }else{
1917
                int j;
1918
                assert(IS_SUB_4X4(sub_mb_type));
1919
                for(j=0; j<4; j++){
1920
                    int sub_x_offset= x_offset + 2*(j&1);
1921
                    int sub_y_offset= y_offset +   (j&2);
1922
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1923
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1924
                        &weight_op[6], &weight_avg[6],
1925
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1926
                }
1927
            }
1928
        }
1929
    }
1930

    
1931
    prefetch_motion(h, 1);
1932
}
1933

    
1934
static av_cold void decode_init_vlc(void){
1935
    static int done = 0;
1936

    
1937
    if (!done) {
1938
        int i;
1939
        int offset;
1940
        done = 1;
1941

    
1942
        chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
1943
        chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
1944
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1945
                 &chroma_dc_coeff_token_len [0], 1, 1,
1946
                 &chroma_dc_coeff_token_bits[0], 1, 1,
1947
                 INIT_VLC_USE_NEW_STATIC);
1948

    
1949
        offset = 0;
1950
        for(i=0; i<4; i++){
1951
            coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
1952
            coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
1953
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1954
                     &coeff_token_len [i][0], 1, 1,
1955
                     &coeff_token_bits[i][0], 1, 1,
1956
                     INIT_VLC_USE_NEW_STATIC);
1957
            offset += coeff_token_vlc_tables_size[i];
1958
        }
1959
        /*
1960
         * This is a one time safety check to make sure that
1961
         * the packed static coeff_token_vlc table sizes
1962
         * were initialized correctly.
1963
         */
1964
        assert(offset == sizeof(coeff_token_vlc_tables)/(sizeof(VLC_TYPE)*2));
1965

    
1966
        for(i=0; i<3; i++){
1967
            chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
1968
            chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
1969
            init_vlc(&chroma_dc_total_zeros_vlc[i],
1970
                     CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1971
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
1972
                     &chroma_dc_total_zeros_bits[i][0], 1, 1,
1973
                     INIT_VLC_USE_NEW_STATIC);
1974
        }
1975
        for(i=0; i<15; i++){
1976
            total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
1977
            total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
1978
            init_vlc(&total_zeros_vlc[i],
1979
                     TOTAL_ZEROS_VLC_BITS, 16,
1980
                     &total_zeros_len [i][0], 1, 1,
1981
                     &total_zeros_bits[i][0], 1, 1,
1982
                     INIT_VLC_USE_NEW_STATIC);
1983
        }
1984

    
1985
        for(i=0; i<6; i++){
1986
            run_vlc[i].table = run_vlc_tables[i];
1987
            run_vlc[i].table_allocated = run_vlc_tables_size;
1988
            init_vlc(&run_vlc[i],
1989
                     RUN_VLC_BITS, 7,
1990
                     &run_len [i][0], 1, 1,
1991
                     &run_bits[i][0], 1, 1,
1992
                     INIT_VLC_USE_NEW_STATIC);
1993
        }
1994
        run7_vlc.table = run7_vlc_table,
1995
        run7_vlc.table_allocated = run7_vlc_table_size;
1996
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1997
                 &run_len [6][0], 1, 1,
1998
                 &run_bits[6][0], 1, 1,
1999
                 INIT_VLC_USE_NEW_STATIC);
2000
    }
2001
}
2002

    
2003
static void free_tables(H264Context *h){
2004
    int i;
2005
    H264Context *hx;
2006
    av_freep(&h->intra4x4_pred_mode);
2007
    av_freep(&h->chroma_pred_mode_table);
2008
    av_freep(&h->cbp_table);
2009
    av_freep(&h->mvd_table[0]);
2010
    av_freep(&h->mvd_table[1]);
2011
    av_freep(&h->direct_table);
2012
    av_freep(&h->non_zero_count);
2013
    av_freep(&h->slice_table_base);
2014
    h->slice_table= NULL;
2015

    
2016
    av_freep(&h->mb2b_xy);
2017
    av_freep(&h->mb2b8_xy);
2018

    
2019
    for(i = 0; i < MAX_SPS_COUNT; i++)
2020
        av_freep(h->sps_buffers + i);
2021

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

    
2025
    for(i = 0; i < h->s.avctx->thread_count; i++) {
2026
        hx = h->thread_context[i];
2027
        if(!hx) continue;
2028
        av_freep(&hx->top_borders[1]);
2029
        av_freep(&hx->top_borders[0]);
2030
        av_freep(&hx->s.obmc_scratchpad);
2031
    }
2032
}
2033

    
2034
static void init_dequant8_coeff_table(H264Context *h){
2035
    int i,q,x;
2036
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2037
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
2038
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
2039

    
2040
    for(i=0; i<2; i++ ){
2041
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2042
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
2043
            break;
2044
        }
2045

    
2046
        for(q=0; q<52; q++){
2047
            int shift = ff_div6[q];
2048
            int idx = ff_rem6[q];
2049
            for(x=0; x<64; x++)
2050
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2051
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2052
                    h->pps.scaling_matrix8[i][x]) << shift;
2053
        }
2054
    }
2055
}
2056

    
2057
static void init_dequant4_coeff_table(H264Context *h){
2058
    int i,j,q,x;
2059
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2060
    for(i=0; i<6; i++ ){
2061
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
2062
        for(j=0; j<i; j++){
2063
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2064
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2065
                break;
2066
            }
2067
        }
2068
        if(j<i)
2069
            continue;
2070

    
2071
        for(q=0; q<52; q++){
2072
            int shift = ff_div6[q] + 2;
2073
            int idx = ff_rem6[q];
2074
            for(x=0; x<16; x++)
2075
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2076
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2077
                    h->pps.scaling_matrix4[i][x]) << shift;
2078
        }
2079
    }
2080
}
2081

    
2082
static void init_dequant_tables(H264Context *h){
2083
    int i,x;
2084
    init_dequant4_coeff_table(h);
2085
    if(h->pps.transform_8x8_mode)
2086
        init_dequant8_coeff_table(h);
2087
    if(h->sps.transform_bypass){
2088
        for(i=0; i<6; i++)
2089
            for(x=0; x<16; x++)
2090
                h->dequant4_coeff[i][0][x] = 1<<6;
2091
        if(h->pps.transform_8x8_mode)
2092
            for(i=0; i<2; i++)
2093
                for(x=0; x<64; x++)
2094
                    h->dequant8_coeff[i][0][x] = 1<<6;
2095
    }
2096
}
2097

    
2098

    
2099
/**
2100
 * allocates tables.
2101
 * needs width/height
2102
 */
2103
static int alloc_tables(H264Context *h){
2104
    MpegEncContext * const s = &h->s;
2105
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2106
    int x,y;
2107

    
2108
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2109

    
2110
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2111
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
2112
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2113

    
2114
    CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2115
    CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2116
    CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2117
    CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2118

    
2119
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));
2120
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2121

    
2122
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2123
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2124
    for(y=0; y<s->mb_height; y++){
2125
        for(x=0; x<s->mb_width; x++){
2126
            const int mb_xy= x + y*s->mb_stride;
2127
            const int b_xy = 4*x + 4*y*h->b_stride;
2128
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2129

    
2130
            h->mb2b_xy [mb_xy]= b_xy;
2131
            h->mb2b8_xy[mb_xy]= b8_xy;
2132
        }
2133
    }
2134

    
2135
    s->obmc_scratchpad = NULL;
2136

    
2137
    if(!h->dequant4_coeff[0])
2138
        init_dequant_tables(h);
2139

    
2140
    return 0;
2141
fail:
2142
    free_tables(h);
2143
    return -1;
2144
}
2145

    
2146
/**
2147
 * Mimic alloc_tables(), but for every context thread.
2148
 */
2149
static void clone_tables(H264Context *dst, H264Context *src){
2150
    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
2151
    dst->non_zero_count           = src->non_zero_count;
2152
    dst->slice_table              = src->slice_table;
2153
    dst->cbp_table                = src->cbp_table;
2154
    dst->mb2b_xy                  = src->mb2b_xy;
2155
    dst->mb2b8_xy                 = src->mb2b8_xy;
2156
    dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
2157
    dst->mvd_table[0]             = src->mvd_table[0];
2158
    dst->mvd_table[1]             = src->mvd_table[1];
2159
    dst->direct_table             = src->direct_table;
2160

    
2161
    dst->s.obmc_scratchpad = NULL;
2162
    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2163
}
2164

    
2165
/**
2166
 * Init context
2167
 * Allocate buffers which are not shared amongst multiple threads.
2168
 */
2169
static int context_init(H264Context *h){
2170
    CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2171
    CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2172

    
2173
    return 0;
2174
fail:
2175
    return -1; // free_tables will clean up for us
2176
}
2177

    
2178
static av_cold void common_init(H264Context *h){
2179
    MpegEncContext * const s = &h->s;
2180

    
2181
    s->width = s->avctx->width;
2182
    s->height = s->avctx->height;
2183
    s->codec_id= s->avctx->codec->id;
2184

    
2185
    ff_h264_pred_init(&h->hpc, s->codec_id);
2186

    
2187
    h->dequant_coeff_pps= -1;
2188
    s->unrestricted_mv=1;
2189
    s->decode=1; //FIXME
2190

    
2191
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2192
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2193
}
2194

    
2195
static av_cold int decode_init(AVCodecContext *avctx){
2196
    H264Context *h= avctx->priv_data;
2197
    MpegEncContext * const s = &h->s;
2198

    
2199
    MPV_decode_defaults(s);
2200

    
2201
    s->avctx = avctx;
2202
    common_init(h);
2203

    
2204
    s->out_format = FMT_H264;
2205
    s->workaround_bugs= avctx->workaround_bugs;
2206

    
2207
    // set defaults
2208
//    s->decode_mb= ff_h263_decode_mb;
2209
    s->quarter_sample = 1;
2210
    s->low_delay= 1;
2211

    
2212
    if(avctx->codec_id == CODEC_ID_SVQ3)
2213
        avctx->pix_fmt= PIX_FMT_YUVJ420P;
2214
    else
2215
        avctx->pix_fmt= PIX_FMT_YUV420P;
2216

    
2217
    decode_init_vlc();
2218

    
2219
    if(avctx->extradata_size > 0 && avctx->extradata &&
2220
       *(char *)avctx->extradata == 1){
2221
        h->is_avc = 1;
2222
        h->got_avcC = 0;
2223
    } else {
2224
        h->is_avc = 0;
2225
    }
2226

    
2227
    h->thread_context[0] = h;
2228
    h->outputed_poc = INT_MIN;
2229
    h->prev_poc_msb= 1<<16;
2230
    return 0;
2231
}
2232

    
2233
static int frame_start(H264Context *h){
2234
    MpegEncContext * const s = &h->s;
2235
    int i;
2236

    
2237
    if(MPV_frame_start(s, s->avctx) < 0)
2238
        return -1;
2239
    ff_er_frame_start(s);
2240
    /*
2241
     * MPV_frame_start uses pict_type to derive key_frame.
2242
     * This is incorrect for H.264; IDR markings must be used.
2243
     * Zero here; IDR markings per slice in frame or fields are ORed in later.
2244
     * See decode_nal_units().
2245
     */
2246
    s->current_picture_ptr->key_frame= 0;
2247

    
2248
    assert(s->linesize && s->uvlinesize);
2249

    
2250
    for(i=0; i<16; i++){
2251
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2252
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2253
    }
2254
    for(i=0; i<4; i++){
2255
        h->block_offset[16+i]=
2256
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2257
        h->block_offset[24+16+i]=
2258
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2259
    }
2260

    
2261
    /* can't be in alloc_tables because linesize isn't known there.
2262
     * FIXME: redo bipred weight to not require extra buffer? */
2263
    for(i = 0; i < s->avctx->thread_count; i++)
2264
        if(!h->thread_context[i]->s.obmc_scratchpad)
2265
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2266

    
2267
    /* some macroblocks will be accessed before they're available */
2268
    if(FRAME_MBAFF || s->avctx->thread_count > 1)
2269
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
2270

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

    
2273
    // We mark the current picture as non-reference after allocating it, so
2274
    // that if we break out due to an error it can be released automatically
2275
    // in the next MPV_frame_start().
2276
    // SVQ3 as well as most other codecs have only last/next/current and thus
2277
    // get released even with set reference, besides SVQ3 and others do not
2278
    // mark frames as reference later "naturally".
2279
    if(s->codec_id != CODEC_ID_SVQ3)
2280
        s->current_picture_ptr->reference= 0;
2281

    
2282
    s->current_picture_ptr->field_poc[0]=
2283
    s->current_picture_ptr->field_poc[1]= INT_MAX;
2284
    assert(s->current_picture_ptr->long_ref==0);
2285

    
2286
    return 0;
2287
}
2288

    
2289
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){
2290
    MpegEncContext * const s = &h->s;
2291
    int i;
2292
    int step    = 1;
2293
    int offset  = 1;
2294
    int uvoffset= 1;
2295
    int top_idx = 1;
2296
    int skiplast= 0;
2297

    
2298
    src_y  -=   linesize;
2299
    src_cb -= uvlinesize;
2300
    src_cr -= uvlinesize;
2301

    
2302
    if(!simple && FRAME_MBAFF){
2303
        if(s->mb_y&1){
2304
            offset  = MB_MBAFF ? 1 : 17;
2305
            uvoffset= MB_MBAFF ? 1 : 9;
2306
            if(!MB_MBAFF){
2307
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y +  15*linesize);
2308
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
2309
                if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2310
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
2311
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
2312
                }
2313
            }
2314
        }else{
2315
            if(!MB_MBAFF){
2316
                h->left_border[0]= h->top_borders[0][s->mb_x][15];
2317
                if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2318
                    h->left_border[34   ]= h->top_borders[0][s->mb_x][16+7  ];
2319
                    h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
2320
                }
2321
                skiplast= 1;
2322
            }
2323
            offset  =
2324
            uvoffset=
2325
            top_idx = MB_MBAFF ? 0 : 1;
2326
        }
2327
        step= MB_MBAFF ? 2 : 1;
2328
    }
2329

    
2330
    // There are two lines saved, the line above the the top macroblock of a pair,
2331
    // and the line above the bottom macroblock
2332
    h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
2333
    for(i=1; i<17 - skiplast; i++){
2334
        h->left_border[offset+i*step]= src_y[15+i*  linesize];
2335
    }
2336

    
2337
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2338
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2339

    
2340
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2341
        h->left_border[uvoffset+34   ]= h->top_borders[top_idx][s->mb_x][16+7];
2342
        h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
2343
        for(i=1; i<9 - skiplast; i++){
2344
            h->left_border[uvoffset+34   +i*step]= src_cb[7+i*uvlinesize];
2345
            h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
2346
        }
2347
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2348
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2349
    }
2350
}
2351

    
2352
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){
2353
    MpegEncContext * const s = &h->s;
2354
    int temp8, i;
2355
    uint64_t temp64;
2356
    int deblock_left;
2357
    int deblock_top;
2358
    int mb_xy;
2359
    int step    = 1;
2360
    int offset  = 1;
2361
    int uvoffset= 1;
2362
    int top_idx = 1;
2363

    
2364
    if(!simple && FRAME_MBAFF){
2365
        if(s->mb_y&1){
2366
            offset  = MB_MBAFF ? 1 : 17;
2367
            uvoffset= MB_MBAFF ? 1 : 9;
2368
        }else{
2369
            offset  =
2370
            uvoffset=
2371
            top_idx = MB_MBAFF ? 0 : 1;
2372
        }
2373
        step= MB_MBAFF ? 2 : 1;
2374
    }
2375

    
2376
    if(h->deblocking_filter == 2) {
2377
        mb_xy = h->mb_xy;
2378
        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2379
        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2380
    } else {
2381
        deblock_left = (s->mb_x > 0);
2382
        deblock_top =  (s->mb_y > 0);
2383
    }
2384

    
2385
    src_y  -=   linesize + 1;
2386
    src_cb -= uvlinesize + 1;
2387
    src_cr -= uvlinesize + 1;
2388

    
2389
#define XCHG(a,b,t,xchg)\
2390
t= a;\
2391
if(xchg)\
2392
    a= b;\
2393
b= t;
2394

    
2395
    if(deblock_left){
2396
        for(i = !deblock_top; i<16; i++){
2397
            XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, xchg);
2398
        }
2399
        XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, 1);
2400
    }
2401

    
2402
    if(deblock_top){
2403
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2404
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2405
        if(s->mb_x+1 < s->mb_width){
2406
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2407
        }
2408
    }
2409

    
2410
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2411
        if(deblock_left){
2412
            for(i = !deblock_top; i<8; i++){
2413
                XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, xchg);
2414
                XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
2415
            }
2416
            XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, 1);
2417
            XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
2418
        }
2419
        if(deblock_top){
2420
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2421
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2422
        }
2423
    }
2424
}
2425

    
2426
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2427
    MpegEncContext * const s = &h->s;
2428
    const int mb_x= s->mb_x;
2429
    const int mb_y= s->mb_y;
2430
    const int mb_xy= h->mb_xy;
2431
    const int mb_type= s->current_picture.mb_type[mb_xy];
2432
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2433
    int linesize, uvlinesize /*dct_offset*/;
2434
    int i;
2435
    int *block_offset = &h->block_offset[0];
2436
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass), is_h264 = (simple || s->codec_id == CODEC_ID_H264);
2437
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2438
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2439

    
2440
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2441
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2442
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2443

    
2444
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2445
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2446

    
2447
    if (!simple && MB_FIELD) {
2448
        linesize   = h->mb_linesize   = s->linesize * 2;
2449
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2450
        block_offset = &h->block_offset[24];
2451
        if(mb_y&1){ //FIXME move out of this function?
2452
            dest_y -= s->linesize*15;
2453
            dest_cb-= s->uvlinesize*7;
2454
            dest_cr-= s->uvlinesize*7;
2455
        }
2456
        if(FRAME_MBAFF) {
2457
            int list;
2458
            for(list=0; list<h->list_count; list++){
2459
                if(!USES_LIST(mb_type, list))
2460
                    continue;
2461
                if(IS_16X16(mb_type)){
2462
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
2463
                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2464
                }else{
2465
                    for(i=0; i<16; i+=4){
2466
                        int ref = h->ref_cache[list][scan8[i]];
2467
                        if(ref >= 0)
2468
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2469
                    }
2470
                }
2471
            }
2472
        }
2473
    } else {
2474
        linesize   = h->mb_linesize   = s->linesize;
2475
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2476
//        dct_offset = s->linesize * 16;
2477
    }
2478

    
2479
    if(transform_bypass){
2480
        idct_dc_add =
2481
        idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2482
    }else if(IS_8x8DCT(mb_type)){
2483
        idct_dc_add = s->dsp.h264_idct8_dc_add;
2484
        idct_add = s->dsp.h264_idct8_add;
2485
    }else{
2486
        idct_dc_add = s->dsp.h264_idct_dc_add;
2487
        idct_add = s->dsp.h264_idct_add;
2488
    }
2489

    
2490
    if (!simple && IS_INTRA_PCM(mb_type)) {
2491
        for (i=0; i<16; i++) {
2492
            memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
2493
        }
2494
        for (i=0; i<8; i++) {
2495
            memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
2496
            memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
2497
        }
2498
    } else {
2499
        if(IS_INTRA(mb_type)){
2500
            if(h->deblocking_filter)
2501
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2502

    
2503
            if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2504
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2505
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2506
            }
2507

    
2508
            if(IS_INTRA4x4(mb_type)){
2509
                if(simple || !s->encoding){
2510
                    if(IS_8x8DCT(mb_type)){
2511
                        for(i=0; i<16; i+=4){
2512
                            uint8_t * const ptr= dest_y + block_offset[i];
2513
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2514
                            const int nnz = h->non_zero_count_cache[ scan8[i] ];
2515
                            h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2516
                                                   (h->topright_samples_available<<i)&0x4000, linesize);
2517
                            if(nnz){
2518
                                if(nnz == 1 && h->mb[i*16])
2519
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
2520
                                else
2521
                                    idct_add(ptr, h->mb + i*16, linesize);
2522
                            }
2523
                        }
2524
                    }else
2525
                    for(i=0; i<16; i++){
2526
                        uint8_t * const ptr= dest_y + block_offset[i];
2527
                        uint8_t *topright;
2528
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2529
                        int nnz, tr;
2530

    
2531
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2532
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2533
                            assert(mb_y || linesize <= block_offset[i]);
2534
                            if(!topright_avail){
2535
                                tr= ptr[3 - linesize]*0x01010101;
2536
                                topright= (uint8_t*) &tr;
2537
                            }else
2538
                                topright= ptr + 4 - linesize;
2539
                        }else
2540
                            topright= NULL;
2541

    
2542
                        h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2543
                        nnz = h->non_zero_count_cache[ scan8[i] ];
2544
                        if(nnz){
2545
                            if(is_h264){
2546
                                if(nnz == 1 && h->mb[i*16])
2547
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
2548
                                else
2549
                                    idct_add(ptr, h->mb + i*16, linesize);
2550
                            }else
2551
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2552
                        }
2553
                    }
2554
                }
2555
            }else{
2556
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2557
                if(is_h264){
2558
                    if(!transform_bypass)
2559
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
2560
                }else
2561
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2562
            }
2563
            if(h->deblocking_filter)
2564
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2565
        }else if(is_h264){
2566
            hl_motion(h, dest_y, dest_cb, dest_cr,
2567
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2568
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2569
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2570
        }
2571

    
2572

    
2573
        if(!IS_INTRA4x4(mb_type)){
2574
            if(is_h264){
2575
                if(IS_INTRA16x16(mb_type)){
2576
                    for(i=0; i<16; i++){
2577
                        if(h->non_zero_count_cache[ scan8[i] ])
2578
                            idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2579
                        else if(h->mb[i*16])
2580
                            idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2581
                    }
2582
                }else{
2583
                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2584
                    for(i=0; i<16; i+=di){
2585
                        int nnz = h->non_zero_count_cache[ scan8[i] ];
2586
                        if(nnz){
2587
                            if(nnz==1 && h->mb[i*16])
2588
                                idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2589
                            else
2590
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2591
                        }
2592
                    }
2593
                }
2594
            }else{
2595
                for(i=0; i<16; i++){
2596
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2597
                        uint8_t * const ptr= dest_y + block_offset[i];
2598
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2599
                    }
2600
                }
2601
            }
2602
        }
2603

    
2604
        if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2605
            uint8_t *dest[2] = {dest_cb, dest_cr};
2606
            if(transform_bypass){
2607
                idct_add = idct_dc_add = s->dsp.add_pixels4;
2608
            }else{
2609
                idct_add = s->dsp.h264_idct_add;
2610
                idct_dc_add = s->dsp.h264_idct_dc_add;
2611
                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]);
2612
                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]);
2613
            }
2614
            if(is_h264){
2615
                for(i=16; i<16+8; i++){
2616
                    if(h->non_zero_count_cache[ scan8[i] ])
2617
                        idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2618
                    else if(h->mb[i*16])
2619
                        idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2620
                }
2621
            }else{
2622
                for(i=16; i<16+8; i++){
2623
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2624
                        uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
2625
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2626
                    }
2627
                }
2628
            }
2629
        }
2630
    }
2631
    if(h->deblocking_filter) {
2632
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
2633
        fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
2634
        h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2635
        h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2636
        if (!simple && FRAME_MBAFF) {
2637
            filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2638
        } else {
2639
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2640
        }
2641
    }
2642
}
2643

    
2644
/**
2645
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2646
 */
2647
static void hl_decode_mb_simple(H264Context *h){
2648
    hl_decode_mb_internal(h, 1);
2649
}
2650

    
2651
/**
2652
 * Process a macroblock; this handles edge cases, such as interlacing.
2653
 */
2654
static void av_noinline hl_decode_mb_complex(H264Context *h){
2655
    hl_decode_mb_internal(h, 0);
2656
}
2657

    
2658
static void hl_decode_mb(H264Context *h){
2659
    MpegEncContext * const s = &h->s;
2660
    const int mb_xy= h->mb_xy;
2661
    const int mb_type= s->current_picture.mb_type[mb_xy];
2662
    int is_complex = FRAME_MBAFF || MB_FIELD || IS_INTRA_PCM(mb_type) || s->codec_id != CODEC_ID_H264 ||
2663
                    (ENABLE_GRAY && (s->flags&CODEC_FLAG_GRAY)) || (ENABLE_H264_ENCODER && s->encoding) || ENABLE_SMALL;
2664

    
2665
    if(ENABLE_H264_ENCODER && !s->decode)
2666
        return;
2667

    
2668
    if (is_complex)
2669
        hl_decode_mb_complex(h);
2670
    else hl_decode_mb_simple(h);
2671
}
2672

    
2673
static void pic_as_field(Picture *pic, const int parity){
2674
    int i;
2675
    for (i = 0; i < 4; ++i) {
2676
        if (parity == PICT_BOTTOM_FIELD)
2677
            pic->data[i] += pic->linesize[i];
2678
        pic->reference = parity;
2679
        pic->linesize[i] *= 2;
2680
    }
2681
    pic->poc= pic->field_poc[parity == PICT_BOTTOM_FIELD];
2682
}
2683

    
2684
static int split_field_copy(Picture *dest, Picture *src,
2685
                            int parity, int id_add){
2686
    int match = !!(src->reference & parity);
2687

    
2688
    if (match) {
2689
        *dest = *src;
2690
        if(parity != PICT_FRAME){
2691
            pic_as_field(dest, parity);
2692
            dest->pic_id *= 2;
2693
            dest->pic_id += id_add;
2694
        }
2695
    }
2696

    
2697
    return match;
2698
}
2699

    
2700
static int build_def_list(Picture *def, Picture **in, int len, int is_long, int sel){
2701
    int i[2]={0};
2702
    int index=0;
2703

    
2704
    while(i[0]<len || i[1]<len){
2705
        while(i[0]<len && !(in[ i[0] ] && (in[ i[0] ]->reference & sel)))
2706
            i[0]++;
2707
        while(i[1]<len && !(in[ i[1] ] && (in[ i[1] ]->reference & (sel^3))))
2708
            i[1]++;
2709
        if(i[0] < len){
2710
            in[ i[0] ]->pic_id= is_long ? i[0] : in[ i[0] ]->frame_num;
2711
            split_field_copy(&def[index++], in[ i[0]++ ], sel  , 1);
2712
        }
2713
        if(i[1] < len){
2714
            in[ i[1] ]->pic_id= is_long ? i[1] : in[ i[1] ]->frame_num;
2715
            split_field_copy(&def[index++], in[ i[1]++ ], sel^3, 0);
2716
        }
2717
    }
2718

    
2719
    return index;
2720
}
2721

    
2722
static int add_sorted(Picture **sorted, Picture **src, int len, int limit, int dir){
2723
    int i, best_poc;
2724
    int out_i= 0;
2725

    
2726
    for(;;){
2727
        best_poc= dir ? INT_MIN : INT_MAX;
2728

    
2729
        for(i=0; i<len; i++){
2730
            const int poc= src[i]->poc;
2731
            if(((poc > limit) ^ dir) && ((poc < best_poc) ^ dir)){
2732
                best_poc= poc;
2733
                sorted[out_i]= src[i];
2734
            }
2735
        }
2736
        if(best_poc == (dir ? INT_MIN : INT_MAX))
2737
            break;
2738
        limit= sorted[out_i++]->poc - dir;
2739
    }
2740
    return out_i;
2741
}
2742

    
2743
/**
2744
 * fills the default_ref_list.
2745
 */
2746
static int fill_default_ref_list(H264Context *h){
2747
    MpegEncContext * const s = &h->s;
2748
    int i, len;
2749

    
2750
    if(h->slice_type_nos==FF_B_TYPE){
2751
        Picture *sorted[32];
2752
        int cur_poc, list;
2753
        int lens[2];
2754

    
2755
        if(FIELD_PICTURE)
2756
            cur_poc= s->current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
2757
        else
2758
            cur_poc= s->current_picture_ptr->poc;
2759

    
2760
        for(list= 0; list<2; list++){
2761
            len= add_sorted(sorted    , h->short_ref, h->short_ref_count, cur_poc, 1^list);
2762
            len+=add_sorted(sorted+len, h->short_ref, h->short_ref_count, cur_poc, 0^list);
2763
            assert(len<=32);
2764
            len= build_def_list(h->default_ref_list[list]    , sorted     , len, 0, s->picture_structure);
2765
            len+=build_def_list(h->default_ref_list[list]+len, h->long_ref, 16 , 1, s->picture_structure);
2766
            assert(len<=32);
2767

    
2768
            if(len < h->ref_count[list])
2769
                memset(&h->default_ref_list[list][len], 0, sizeof(Picture)*(h->ref_count[list] - len));
2770
            lens[list]= len;
2771
        }
2772

    
2773
        if(lens[0] == lens[1] && lens[1] > 1){
2774
            for(i=0; h->default_ref_list[0][i].data[0] == h->default_ref_list[1][i].data[0] && i<lens[0]; i++);
2775
            if(i == lens[0])
2776
                FFSWAP(Picture, h->default_ref_list[1][0], h->default_ref_list[1][1]);
2777
        }
2778
    }else{
2779
        len = build_def_list(h->default_ref_list[0]    , h->short_ref, h->short_ref_count, 0, s->picture_structure);
2780
        len+= build_def_list(h->default_ref_list[0]+len, h-> long_ref, 16                , 1, s->picture_structure);
2781
        assert(len <= 32);
2782
        if(len < h->ref_count[0])
2783
            memset(&h->default_ref_list[0][len], 0, sizeof(Picture)*(h->ref_count[0] - len));
2784
    }
2785
#ifdef TRACE
2786
    for (i=0; i<h->ref_count[0]; i++) {
2787
        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]);
2788
    }
2789
    if(h->slice_type_nos==FF_B_TYPE){
2790
        for (i=0; i<h->ref_count[1]; i++) {
2791
            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]);
2792
        }
2793
    }
2794
#endif
2795
    return 0;
2796
}
2797

    
2798
static void print_short_term(H264Context *h);
2799
static void print_long_term(H264Context *h);
2800

    
2801
/**
2802
 * Extract structure information about the picture described by pic_num in
2803
 * the current decoding context (frame or field). Note that pic_num is
2804
 * picture number without wrapping (so, 0<=pic_num<max_pic_num).
2805
 * @param pic_num picture number for which to extract structure information
2806
 * @param structure one of PICT_XXX describing structure of picture
2807
 *                      with pic_num
2808
 * @return frame number (short term) or long term index of picture
2809
 *         described by pic_num
2810
 */
2811
static int pic_num_extract(H264Context *h, int pic_num, int *structure){
2812
    MpegEncContext * const s = &h->s;
2813

    
2814
    *structure = s->picture_structure;
2815
    if(FIELD_PICTURE){
2816
        if (!(pic_num & 1))
2817
            /* opposite field */
2818
            *structure ^= PICT_FRAME;
2819
        pic_num >>= 1;
2820
    }
2821

    
2822
    return pic_num;
2823
}
2824

    
2825
static int decode_ref_pic_list_reordering(H264Context *h){
2826
    MpegEncContext * const s = &h->s;
2827
    int list, index, pic_structure;
2828

    
2829
    print_short_term(h);
2830
    print_long_term(h);
2831

    
2832
    for(list=0; list<h->list_count; list++){
2833
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2834

    
2835
        if(get_bits1(&s->gb)){
2836
            int pred= h->curr_pic_num;
2837

    
2838
            for(index=0; ; index++){
2839
                unsigned int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2840
                unsigned int pic_id;
2841
                int i;
2842
                Picture *ref = NULL;
2843

    
2844
                if(reordering_of_pic_nums_idc==3)
2845
                    break;
2846

    
2847
                if(index >= h->ref_count[list]){
2848
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2849
                    return -1;
2850
                }
2851

    
2852
                if(reordering_of_pic_nums_idc<3){
2853
                    if(reordering_of_pic_nums_idc<2){
2854
                        const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2855
                        int frame_num;
2856

    
2857
                        if(abs_diff_pic_num > h->max_pic_num){
2858
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2859
                            return -1;
2860
                        }
2861

    
2862
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2863
                        else                                pred+= abs_diff_pic_num;
2864
                        pred &= h->max_pic_num - 1;
2865

    
2866
                        frame_num = pic_num_extract(h, pred, &pic_structure);
2867

    
2868
                        for(i= h->short_ref_count-1; i>=0; i--){
2869
                            ref = h->short_ref[i];
2870
                            assert(ref->reference);
2871
                            assert(!ref->long_ref);
2872
                            if(
2873
                                   ref->frame_num == frame_num &&
2874
                                   (ref->reference & pic_structure)
2875
                              )
2876
                                break;
2877
                        }
2878
                        if(i>=0)
2879
                            ref->pic_id= pred;
2880
                    }else{
2881
                        int long_idx;
2882
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2883

    
2884
                        long_idx= pic_num_extract(h, pic_id, &pic_structure);
2885

    
2886
                        if(long_idx>31){
2887
                            av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
2888
                            return -1;
2889
                        }
2890
                        ref = h->long_ref[long_idx];
2891
                        assert(!(ref && !ref->reference));
2892
                        if(ref && (ref->reference & pic_structure)){
2893
                            ref->pic_id= pic_id;
2894
                            assert(ref->long_ref);
2895
                            i=0;
2896
                        }else{
2897
                            i=-1;
2898
                        }
2899
                    }
2900

    
2901
                    if (i < 0) {
2902
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2903
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2904
                    } else {
2905
                        for(i=index; i+1<h->ref_count[list]; i++){
2906
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
2907
                                break;
2908
                        }
2909
                        for(; i > index; i--){
2910
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2911
                        }
2912
                        h->ref_list[list][index]= *ref;
2913
                        if (FIELD_PICTURE){
2914
                            pic_as_field(&h->ref_list[list][index], pic_structure);
2915
                        }
2916
                    }
2917
                }else{
2918
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2919
                    return -1;
2920
                }
2921
            }
2922
        }
2923
    }
2924
    for(list=0; list<h->list_count; list++){
2925
        for(index= 0; index < h->ref_count[list]; index++){
2926
            if(!h->ref_list[list][index].data[0]){
2927
                av_log(h->s.avctx, AV_LOG_ERROR, "Missing reference picture\n");
2928
                h->ref_list[list][index]= s->current_picture; //FIXME this is not a sensible solution
2929
            }
2930
        }
2931
    }
2932

    
2933
    return 0;
2934
}
2935

    
2936
static void fill_mbaff_ref_list(H264Context *h){
2937
    int list, i, j;
2938
    for(list=0; list<2; list++){ //FIXME try list_count
2939
        for(i=0; i<h->ref_count[list]; i++){
2940
            Picture *frame = &h->ref_list[list][i];
2941
            Picture *field = &h->ref_list[list][16+2*i];
2942
            field[0] = *frame;
2943
            for(j=0; j<3; j++)
2944
                field[0].linesize[j] <<= 1;
2945
            field[0].reference = PICT_TOP_FIELD;
2946
            field[0].poc= field[0].field_poc[0];
2947
            field[1] = field[0];
2948
            for(j=0; j<3; j++)
2949
                field[1].data[j] += frame->linesize[j];
2950
            field[1].reference = PICT_BOTTOM_FIELD;
2951
            field[1].poc= field[1].field_poc[1];
2952

    
2953
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
2954
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
2955
            for(j=0; j<2; j++){
2956
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
2957
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
2958
            }
2959
        }
2960
    }
2961
    for(j=0; j<h->ref_count[1]; j++){
2962
        for(i=0; i<h->ref_count[0]; i++)
2963
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
2964
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
2965
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
2966
    }
2967
}
2968

    
2969
static int pred_weight_table(H264Context *h){
2970
    MpegEncContext * const s = &h->s;
2971
    int list, i;
2972
    int luma_def, chroma_def;
2973

    
2974
    h->use_weight= 0;
2975
    h->use_weight_chroma= 0;
2976
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2977
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2978
    luma_def = 1<<h->luma_log2_weight_denom;
2979
    chroma_def = 1<<h->chroma_log2_weight_denom;
2980

    
2981
    for(list=0; list<2; list++){
2982
        for(i=0; i<h->ref_count[list]; i++){
2983
            int luma_weight_flag, chroma_weight_flag;
2984

    
2985
            luma_weight_flag= get_bits1(&s->gb);
2986
            if(luma_weight_flag){
2987
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
2988
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
2989
                if(   h->luma_weight[list][i] != luma_def
2990
                   || h->luma_offset[list][i] != 0)
2991
                    h->use_weight= 1;
2992
            }else{
2993
                h->luma_weight[list][i]= luma_def;
2994
                h->luma_offset[list][i]= 0;
2995
            }
2996

    
2997
            if(CHROMA){
2998
                chroma_weight_flag= get_bits1(&s->gb);
2999
                if(chroma_weight_flag){
3000
                    int j;
3001
                    for(j=0; j<2; j++){
3002
                        h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3003
                        h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3004
                        if(   h->chroma_weight[list][i][j] != chroma_def
3005
                        || h->chroma_offset[list][i][j] != 0)
3006
                            h->use_weight_chroma= 1;
3007
                    }
3008
                }else{
3009
                    int j;
3010
                    for(j=0; j<2; j++){
3011
                        h->chroma_weight[list][i][j]= chroma_def;
3012
                        h->chroma_offset[list][i][j]= 0;
3013
                    }
3014
                }
3015
            }
3016
        }
3017
        if(h->slice_type_nos != FF_B_TYPE) break;
3018
    }
3019
    h->use_weight= h->use_weight || h->use_weight_chroma;
3020
    return 0;
3021
}
3022

    
3023
static void implicit_weight_table(H264Context *h){
3024
    MpegEncContext * const s = &h->s;
3025
    int ref0, ref1;
3026
    int cur_poc = s->current_picture_ptr->poc;
3027

    
3028
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3029
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3030
        h->use_weight= 0;
3031
        h->use_weight_chroma= 0;
3032
        return;
3033
    }
3034

    
3035
    h->use_weight= 2;
3036
    h->use_weight_chroma= 2;
3037
    h->luma_log2_weight_denom= 5;
3038
    h->chroma_log2_weight_denom= 5;
3039

    
3040
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3041
        int poc0 = h->ref_list[0][ref0].poc;
3042
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3043
            int poc1 = h->ref_list[1][ref1].poc;
3044
            int td = av_clip(poc1 - poc0, -128, 127);
3045
            if(td){
3046
                int tb = av_clip(cur_poc - poc0, -128, 127);
3047
                int tx = (16384 + (FFABS(td) >> 1)) / td;
3048
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3049
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3050
                    h->implicit_weight[ref0][ref1] = 32;
3051
                else
3052
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3053
            }else
3054
                h->implicit_weight[ref0][ref1] = 32;
3055
        }
3056
    }
3057
}
3058

    
3059
/**
3060
 * Mark a picture as no longer needed for reference. The refmask
3061
 * argument allows unreferencing of individual fields or the whole frame.
3062
 * If the picture becomes entirely unreferenced, but is being held for
3063
 * display purposes, it is marked as such.
3064
 * @param refmask mask of fields to unreference; the mask is bitwise
3065
 *                anded with the reference marking of pic
3066
 * @return non-zero if pic becomes entirely unreferenced (except possibly
3067
 *         for display purposes) zero if one of the fields remains in
3068
 *         reference
3069
 */
3070
static inline int unreference_pic(H264Context *h, Picture *pic, int refmask){
3071
    int i;
3072
    if (pic->reference &= refmask) {
3073
        return 0;
3074
    } else {
3075
        for(i = 0; h->delayed_pic[i]; i++)
3076
            if(pic == h->delayed_pic[i]){
3077
                pic->reference=DELAYED_PIC_REF;
3078
                break;
3079
            }
3080
        return 1;
3081
    }
3082
}
3083

    
3084
/**
3085
 * instantaneous decoder refresh.
3086
 */
3087
static void idr(H264Context *h){
3088
    int i;
3089

    
3090
    for(i=0; i<16; i++){
3091
        remove_long(h, i, 0);
3092
    }
3093
    assert(h->long_ref_count==0);
3094

    
3095
    for(i=0; i<h->short_ref_count; i++){
3096
        unreference_pic(h, h->short_ref[i], 0);
3097
        h->short_ref[i]= NULL;
3098
    }
3099
    h->short_ref_count=0;
3100
    h->prev_frame_num= 0;
3101
    h->prev_frame_num_offset= 0;
3102
    h->prev_poc_msb=
3103
    h->prev_poc_lsb= 0;
3104
}
3105

    
3106
/* forget old pics after a seek */
3107
static void flush_dpb(AVCodecContext *avctx){
3108
    H264Context *h= avctx->priv_data;
3109
    int i;
3110
    for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
3111
        if(h->delayed_pic[i])
3112
            h->delayed_pic[i]->reference= 0;
3113
        h->delayed_pic[i]= NULL;
3114
    }
3115
    h->outputed_poc= INT_MIN;
3116
    idr(h);
3117
    if(h->s.current_picture_ptr)
3118
        h->s.current_picture_ptr->reference= 0;
3119
    h->s.first_field= 0;
3120
    ff_mpeg_flush(avctx);
3121
}
3122

    
3123
/**
3124
 * Find a Picture in the short term reference list by frame number.
3125
 * @param frame_num frame number to search for
3126
 * @param idx the index into h->short_ref where returned picture is found
3127
 *            undefined if no picture found.
3128
 * @return pointer to the found picture, or NULL if no pic with the provided
3129
 *                 frame number is found
3130
 */
3131
static Picture * find_short(H264Context *h, int frame_num, int *idx){
3132
    MpegEncContext * const s = &h->s;
3133
    int i;
3134

    
3135
    for(i=0; i<h->short_ref_count; i++){
3136
        Picture *pic= h->short_ref[i];
3137
        if(s->avctx->debug&FF_DEBUG_MMCO)
3138
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3139
        if(pic->frame_num == frame_num) {
3140
            *idx = i;
3141
            return pic;
3142
        }
3143
    }
3144
    return NULL;
3145
}
3146

    
3147
/**
3148
 * Remove a picture from the short term reference list by its index in
3149
 * that list.  This does no checking on the provided index; it is assumed
3150
 * to be valid. Other list entries are shifted down.
3151
 * @param i index into h->short_ref of picture to remove.
3152
 */
3153
static void remove_short_at_index(H264Context *h, int i){
3154
    assert(i >= 0 && i < h->short_ref_count);
3155
    h->short_ref[i]= NULL;
3156
    if (--h->short_ref_count)
3157
        memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i)*sizeof(Picture*));
3158
}
3159

    
3160
/**
3161
 *
3162
 * @return the removed picture or NULL if an error occurs
3163
 */
3164
static Picture * remove_short(H264Context *h, int frame_num, int ref_mask){
3165
    MpegEncContext * const s = &h->s;
3166
    Picture *pic;
3167
    int i;
3168

    
3169
    if(s->avctx->debug&FF_DEBUG_MMCO)
3170
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3171

    
3172
    pic = find_short(h, frame_num, &i);
3173
    if (pic){
3174
        if(unreference_pic(h, pic, ref_mask))
3175
        remove_short_at_index(h, i);
3176
    }
3177

    
3178
    return pic;
3179
}
3180

    
3181
/**
3182
 * Remove a picture from the long term reference list by its index in
3183
 * that list.
3184
 * @return the removed picture or NULL if an error occurs
3185
 */
3186
static Picture * remove_long(H264Context *h, int i, int ref_mask){
3187
    Picture *pic;
3188

    
3189
    pic= h->long_ref[i];
3190
    if (pic){
3191
        if(unreference_pic(h, pic, ref_mask)){
3192
            assert(h->long_ref[i]->long_ref == 1);
3193
            h->long_ref[i]->long_ref= 0;
3194
            h->long_ref[i]= NULL;
3195
            h->long_ref_count--;
3196
        }
3197
    }
3198

    
3199
    return pic;
3200
}
3201

    
3202
/**
3203
 * print short term list
3204
 */
3205
static void print_short_term(H264Context *h) {
3206
    uint32_t i;
3207
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3208
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3209
        for(i=0; i<h->short_ref_count; i++){
3210
            Picture *pic= h->short_ref[i];
3211
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3212
        }
3213
    }
3214
}
3215

    
3216
/**
3217
 * print long term list
3218
 */
3219
static void print_long_term(H264Context *h) {
3220
    uint32_t i;
3221
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3222
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3223
        for(i = 0; i < 16; i++){
3224
            Picture *pic= h->long_ref[i];
3225
            if (pic) {
3226
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3227
            }
3228
        }
3229
    }
3230
}
3231

    
3232
/**
3233
 * Executes the reference picture marking (memory management control operations).
3234
 */
3235
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3236
    MpegEncContext * const s = &h->s;
3237
    int i, j;
3238
    int current_ref_assigned=0;
3239
    Picture *pic;
3240

    
3241
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3242
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3243

    
3244
    for(i=0; i<mmco_count; i++){
3245
        int structure, frame_num;
3246
        if(s->avctx->debug&FF_DEBUG_MMCO)
3247
            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);
3248

    
3249
        if(   mmco[i].opcode == MMCO_SHORT2UNUSED
3250
           || mmco[i].opcode == MMCO_SHORT2LONG){
3251
            frame_num = pic_num_extract(h, mmco[i].short_pic_num, &structure);
3252
            pic = find_short(h, frame_num, &j);
3253
            if(!pic){
3254
                if(mmco[i].opcode != MMCO_SHORT2LONG || !h->long_ref[mmco[i].long_arg]
3255
                   || h->long_ref[mmco[i].long_arg]->frame_num != frame_num)
3256
                av_log(h->s.avctx, AV_LOG_ERROR, "mmco: unref short failure\n");
3257
                continue;
3258
            }
3259
        }
3260

    
3261
        switch(mmco[i].opcode){
3262
        case MMCO_SHORT2UNUSED:
3263
            if(s->avctx->debug&FF_DEBUG_MMCO)
3264
                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);
3265
            remove_short(h, frame_num, structure ^ PICT_FRAME);
3266
            break;
3267
        case MMCO_SHORT2LONG:
3268
                if (h->long_ref[mmco[i].long_arg] != pic)
3269
                    remove_long(h, mmco[i].long_arg, 0);
3270

    
3271
                remove_short_at_index(h, j);
3272
                h->long_ref[ mmco[i].long_arg ]= pic;
3273
                if (h->long_ref[ mmco[i].long_arg ]){
3274
                    h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3275
                    h->long_ref_count++;
3276
                }
3277
            break;
3278
        case MMCO_LONG2UNUSED:
3279
            j = pic_num_extract(h, mmco[i].long_arg, &structure);
3280
            pic = h->long_ref[j];
3281
            if (pic) {
3282
                remove_long(h, j, structure ^ PICT_FRAME);
3283
            } else if(s->avctx->debug&FF_DEBUG_MMCO)
3284
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref long failure\n");
3285
            break;
3286
        case MMCO_LONG:
3287
                    // Comment below left from previous code as it is an interresting note.
3288
                    /* First field in pair is in short term list or
3289
                     * at a different long term index.
3290
                     * This is not allowed; see 7.4.3.3, notes 2 and 3.
3291
                     * Report the problem and keep the pair where it is,
3292
                     * and mark this field valid.
3293
                     */
3294

    
3295
            if (h->long_ref[mmco[i].long_arg] != s->current_picture_ptr) {
3296
                remove_long(h, mmco[i].long_arg, 0);
3297

    
3298
                h->long_ref[ mmco[i].long_arg ]= s->current_picture_ptr;
3299
                h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3300
                h->long_ref_count++;
3301
            }
3302

    
3303
            s->current_picture_ptr->reference |= s->picture_structure;
3304
            current_ref_assigned=1;
3305
            break;
3306
        case MMCO_SET_MAX_LONG:
3307
            assert(mmco[i].long_arg <= 16);
3308
            // just remove the long term which index is greater than new max
3309
            for(j = mmco[i].long_arg; j<16; j++){
3310
                remove_long(h, j, 0);
3311
            }
3312
            break;
3313
        case MMCO_RESET:
3314
            while(h->short_ref_count){
3315
                remove_short(h, h->short_ref[0]->frame_num, 0);
3316
            }
3317
            for(j = 0; j < 16; j++) {
3318
                remove_long(h, j, 0);
3319
            }
3320
            s->current_picture_ptr->poc=
3321
            s->current_picture_ptr->field_poc[0]=
3322
            s->current_picture_ptr->field_poc[1]=
3323
            h->poc_lsb=
3324
            h->poc_msb=
3325
            h->frame_num=
3326
            s->current_picture_ptr->frame_num= 0;
3327
            break;
3328
        default: assert(0);
3329
        }
3330
    }
3331

    
3332
    if (!current_ref_assigned) {
3333
        /* Second field of complementary field pair; the first field of
3334
         * which is already referenced. If short referenced, it
3335
         * should be first entry in short_ref. If not, it must exist
3336
         * in long_ref; trying to put it on the short list here is an
3337
         * error in the encoded bit stream (ref: 7.4.3.3, NOTE 2 and 3).
3338
         */
3339
        if (h->short_ref_count && h->short_ref[0] == s->current_picture_ptr) {
3340
            /* Just mark the second field valid */
3341
            s->current_picture_ptr->reference = PICT_FRAME;
3342
        } else if (s->current_picture_ptr->long_ref) {
3343
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term reference "
3344
                                             "assignment for second field "
3345
                                             "in complementary field pair "
3346
                                             "(first field is long term)\n");
3347
        } else {
3348
            pic= remove_short(h, s->current_picture_ptr->frame_num, 0);
3349
            if(pic){
3350
                av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3351
            }
3352

    
3353
            if(h->short_ref_count)
3354
                memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3355

    
3356
            h->short_ref[0]= s->current_picture_ptr;
3357
            h->short_ref_count++;
3358
            s->current_picture_ptr->reference |= s->picture_structure;
3359
        }
3360
    }
3361

    
3362
    if (h->long_ref_count + h->short_ref_count > h->sps.ref_frame_count){
3363

    
3364
        /* We have too many reference frames, probably due to corrupted
3365
         * stream. Need to discard one frame. Prevents overrun of the
3366
         * short_ref and long_ref buffers.
3367
         */
3368
        av_log(h->s.avctx, AV_LOG_ERROR,
3369
               "number of reference frames exceeds max (probably "
3370
               "corrupt input), discarding one\n");
3371

    
3372
        if (h->long_ref_count && !h->short_ref_count) {
3373
            for (i = 0; i < 16; ++i)
3374
                if (h->long_ref[i])
3375
                    break;
3376

    
3377
            assert(i < 16);
3378
            remove_long(h, i, 0);
3379
        } else {
3380
            pic = h->short_ref[h->short_ref_count - 1];
3381
            remove_short(h, pic->frame_num, 0);
3382
        }
3383
    }
3384

    
3385
    print_short_term(h);
3386
    print_long_term(h);
3387
    return 0;
3388
}
3389

    
3390
static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
3391
    MpegEncContext * const s = &h->s;
3392
    int i;
3393

    
3394
    h->mmco_index= 0;
3395
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3396
        s->broken_link= get_bits1(gb) -1;
3397
        if(get_bits1(gb)){
3398
            h->mmco[0].opcode= MMCO_LONG;
3399
            h->mmco[0].long_arg= 0;
3400
            h->mmco_index= 1;
3401
        }
3402
    }else{
3403
        if(get_bits1(gb)){ // adaptive_ref_pic_marking_mode_flag
3404
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
3405
                MMCOOpcode opcode= get_ue_golomb(gb);
3406

    
3407
                h->mmco[i].opcode= opcode;
3408
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3409
                    h->mmco[i].short_pic_num= (h->curr_pic_num - get_ue_golomb(gb) - 1) & (h->max_pic_num - 1);
3410
/*                    if(h->mmco[i].short_pic_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_pic_num ] == NULL){
3411
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
3412
                        return -1;
3413
                    }*/
3414
                }
3415
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3416
                    unsigned int long_arg= get_ue_golomb(gb);
3417
                    if(long_arg >= 32 || (long_arg >= 16 && !(opcode == MMCO_LONG2UNUSED && FIELD_PICTURE))){
3418
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3419
                        return -1;
3420
                    }
3421
                    h->mmco[i].long_arg= long_arg;
3422
                }
3423

    
3424
                if(opcode > (unsigned)MMCO_LONG){
3425
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3426
                    return -1;
3427
                }
3428
                if(opcode == MMCO_END)
3429
                    break;
3430
            }
3431
            h->mmco_index= i;
3432
        }else{
3433
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3434

    
3435
            if(h->short_ref_count && h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count &&
3436
                    !(FIELD_PICTURE && !s->first_field && s->current_picture_ptr->reference)) {
3437
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3438
                h->mmco[0].short_pic_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3439
                h->mmco_index= 1;
3440
                if (FIELD_PICTURE) {
3441
                    h->mmco[0].short_pic_num *= 2;
3442
                    h->mmco[1].opcode= MMCO_SHORT2UNUSED;
3443
                    h->mmco[1].short_pic_num= h->mmco[0].short_pic_num + 1;
3444
                    h->mmco_index= 2;
3445
                }
3446
            }
3447
        }
3448
    }
3449

    
3450
    return 0;
3451
}
3452

    
3453
static int init_poc(H264Context *h){
3454
    MpegEncContext * const s = &h->s;
3455
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3456
    int field_poc[2];
3457
    Picture *cur = s->current_picture_ptr;
3458

    
3459
    h->frame_num_offset= h->prev_frame_num_offset;
3460
    if(h->frame_num < h->prev_frame_num)
3461
        h->frame_num_offset += max_frame_num;
3462

    
3463
    if(h->sps.poc_type==0){
3464
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3465

    
3466
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3467
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3468
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3469
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3470
        else
3471
            h->poc_msb = h->prev_poc_msb;
3472
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3473
        field_poc[0] =
3474
        field_poc[1] = h->poc_msb + h->poc_lsb;
3475
        if(s->picture_structure == PICT_FRAME)
3476
            field_poc[1] += h->delta_poc_bottom;
3477
    }else if(h->sps.poc_type==1){
3478
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3479
        int i;
3480

    
3481
        if(h->sps.poc_cycle_length != 0)
3482
            abs_frame_num = h->frame_num_offset + h->frame_num;
3483
        else
3484
            abs_frame_num = 0;
3485

    
3486
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
3487
            abs_frame_num--;
3488

    
3489
        expected_delta_per_poc_cycle = 0;
3490
        for(i=0; i < h->sps.poc_cycle_length; i++)
3491
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3492

    
3493
        if(abs_frame_num > 0){
3494
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3495
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3496

    
3497
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3498
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
3499
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3500
        } else
3501
            expectedpoc = 0;
3502

    
3503
        if(h->nal_ref_idc == 0)
3504
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3505

    
3506
        field_poc[0] = expectedpoc + h->delta_poc[0];
3507
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3508

    
3509
        if(s->picture_structure == PICT_FRAME)
3510
            field_poc[1] += h->delta_poc[1];
3511
    }else{
3512
        int poc= 2*(h->frame_num_offset + h->frame_num);
3513

    
3514
        if(!h->nal_ref_idc)
3515
            poc--;
3516

    
3517
        field_poc[0]= poc;
3518
        field_poc[1]= poc;
3519
    }
3520

    
3521
    if(s->picture_structure != PICT_BOTTOM_FIELD)
3522
        s->current_picture_ptr->field_poc[0]= field_poc[0];
3523
    if(s->picture_structure != PICT_TOP_FIELD)
3524
        s->current_picture_ptr->field_poc[1]= field_poc[1];
3525
    cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
3526

    
3527
    return 0;
3528
}
3529

    
3530

    
3531
/**
3532
 * initialize scan tables
3533
 */
3534
static void init_scan_tables(H264Context *h){
3535
    MpegEncContext * const s = &h->s;
3536
    int i;
3537
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
3538
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
3539
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
3540
    }else{
3541
        for(i=0; i<16; i++){
3542
#define T(x) (x>>2) | ((x<<2) & 0xF)
3543
            h->zigzag_scan[i] = T(zigzag_scan[i]);
3544
            h-> field_scan[i] = T( field_scan[i]);
3545
#undef T
3546
        }
3547
    }
3548
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
3549
        memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));
3550
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
3551
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
3552
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
3553
    }else{
3554
        for(i=0; i<64; i++){
3555
#define T(x) (x>>3) | ((x&7)<<3)
3556
            h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);
3557
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
3558
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
3559
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
3560
#undef T
3561
        }
3562
    }
3563
    if(h->sps.transform_bypass){ //FIXME same ugly
3564
        h->zigzag_scan_q0          = zigzag_scan;
3565
        h->zigzag_scan8x8_q0       = zigzag_scan8x8;
3566
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
3567
        h->field_scan_q0           = field_scan;
3568
        h->field_scan8x8_q0        = field_scan8x8;
3569
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
3570
    }else{
3571
        h->zigzag_scan_q0          = h->zigzag_scan;
3572
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
3573
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
3574
        h->field_scan_q0           = h->field_scan;
3575
        h->field_scan8x8_q0        = h->field_scan8x8;
3576
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
3577
    }
3578
}
3579

    
3580
/**
3581
 * Replicates H264 "master" context to thread contexts.
3582
 */
3583
static void clone_slice(H264Context *dst, H264Context *src)
3584
{
3585
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
3586
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
3587
    dst->s.current_picture      = src->s.current_picture;
3588
    dst->s.linesize             = src->s.linesize;
3589
    dst->s.uvlinesize           = src->s.uvlinesize;
3590
    dst->s.first_field          = src->s.first_field;
3591

    
3592
    dst->prev_poc_msb           = src->prev_poc_msb;
3593
    dst->prev_poc_lsb           = src->prev_poc_lsb;
3594
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
3595
    dst->prev_frame_num         = src->prev_frame_num;
3596
    dst->short_ref_count        = src->short_ref_count;
3597

    
3598
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
3599
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
3600
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
3601
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
3602

    
3603
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
3604
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
3605
}
3606

    
3607
/**
3608
 * decodes a slice header.
3609
 * This will also call MPV_common_init() and frame_start() as needed.
3610
 *
3611
 * @param h h264context
3612
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
3613
 *
3614
 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3615
 */
3616
static int decode_slice_header(H264Context *h, H264Context *h0){
3617
    MpegEncContext * const s = &h->s;
3618
    MpegEncContext * const s0 = &h0->s;
3619
    unsigned int first_mb_in_slice;
3620
    unsigned int pps_id;
3621
    int num_ref_idx_active_override_flag;
3622
    static const uint8_t slice_type_map[5]= {FF_P_TYPE, FF_B_TYPE, FF_I_TYPE, FF_SP_TYPE, FF_SI_TYPE};
3623
    unsigned int slice_type, tmp, i, j;
3624
    int default_ref_list_done = 0;
3625
    int last_pic_structure;
3626

    
3627
    s->dropable= h->nal_ref_idc == 0;
3628

    
3629
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
3630
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
3631
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
3632
    }else{
3633
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
3634
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
3635
    }
3636

    
3637
    first_mb_in_slice= get_ue_golomb(&s->gb);
3638

    
3639
    if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
3640
        h0->current_slice = 0;
3641
        if (!s0->first_field)
3642
            s->current_picture_ptr= NULL;
3643
    }
3644

    
3645
    slice_type= get_ue_golomb(&s->gb);
3646
    if(slice_type > 9){
3647
        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);
3648
        return -1;
3649
    }
3650
    if(slice_type > 4){
3651
        slice_type -= 5;
3652
        h->slice_type_fixed=1;
3653
    }else
3654
        h->slice_type_fixed=0;
3655

    
3656
    slice_type= slice_type_map[ slice_type ];
3657
    if (slice_type == FF_I_TYPE
3658
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
3659
        default_ref_list_done = 1;
3660
    }
3661
    h->slice_type= slice_type;
3662
    h->slice_type_nos= slice_type & 3;
3663

    
3664
    s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
3665
    if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
3666
        av_log(h->s.avctx, AV_LOG_ERROR,
3667
               "B picture before any references, skipping\n");
3668
        return -1;
3669
    }
3670

    
3671
    pps_id= get_ue_golomb(&s->gb);
3672
    if(pps_id>=MAX_PPS_COUNT){
3673
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3674
        return -1;
3675
    }
3676
    if(!h0->pps_buffers[pps_id]) {
3677
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS referenced\n");
3678
        return -1;
3679
    }
3680
    h->pps= *h0->pps_buffers[pps_id];
3681

    
3682
    if(!h0->sps_buffers[h->pps.sps_id]) {
3683
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS referenced\n");
3684
        return -1;
3685
    }
3686
    h->sps = *h0->sps_buffers[h->pps.sps_id];
3687

    
3688
    if(h == h0 && h->dequant_coeff_pps != pps_id){
3689
        h->dequant_coeff_pps = pps_id;
3690
        init_dequant_tables(h);
3691
    }
3692

    
3693
    s->mb_width= h->sps.mb_width;
3694
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3695

    
3696
    h->b_stride=  s->mb_width*4;
3697
    h->b8_stride= s->mb_width*2;
3698

    
3699
    s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
3700
    if(h->sps.frame_mbs_only_flag)
3701
        s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
3702
    else
3703
        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
3704

    
3705
    if (s->context_initialized
3706
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3707
        if(h != h0)
3708
            return -1;   // width / height changed during parallelized decoding
3709
        free_tables(h);
3710
        MPV_common_end(s);
3711
    }
3712
    if (!s->context_initialized) {
3713
        if(h != h0)
3714
            return -1;  // we cant (re-)initialize context during parallel decoding
3715
        if (MPV_common_init(s) < 0)
3716
            return -1;
3717
        s->first_field = 0;
3718

    
3719
        init_scan_tables(h);
3720
        alloc_tables(h);
3721

    
3722
        for(i = 1; i < s->avctx->thread_count; i++) {
3723
            H264Context *c;
3724
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
3725
            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
3726
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
3727
            c->sps = h->sps;
3728
            c->pps = h->pps;
3729
            init_scan_tables(c);
3730
            clone_tables(c, h);
3731
        }
3732

    
3733
        for(i = 0; i < s->avctx->thread_count; i++)
3734
            if(context_init(h->thread_context[i]) < 0)
3735
                return -1;
3736

    
3737
        s->avctx->width = s->width;
3738
        s->avctx->height = s->height;
3739
        s->avctx->sample_aspect_ratio= h->sps.sar;
3740
        if(!s->avctx->sample_aspect_ratio.den)
3741
            s->avctx->sample_aspect_ratio.den = 1;
3742

    
3743
        if(h->sps.timing_info_present_flag){
3744
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale};
3745
            if(h->x264_build > 0 && h->x264_build < 44)
3746
                s->avctx->time_base.den *= 2;
3747
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3748
                      s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
3749
        }
3750
    }
3751

    
3752
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3753

    
3754
    h->mb_mbaff = 0;
3755
    h->mb_aff_frame = 0;
3756
    last_pic_structure = s0->picture_structure;
3757
    if(h->sps.frame_mbs_only_flag){
3758
        s->picture_structure= PICT_FRAME;
3759
    }else{
3760
        if(get_bits1(&s->gb)) { //field_pic_flag
3761
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3762
        } else {
3763
            s->picture_structure= PICT_FRAME;
3764
            h->mb_aff_frame = h->sps.mb_aff;
3765
        }
3766
    }
3767
    h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
3768

    
3769
    if(h0->current_slice == 0){
3770
        while(h->frame_num !=  h->prev_frame_num &&
3771
              h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
3772
            av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
3773
            frame_start(h);
3774
            h->prev_frame_num++;
3775
            h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
3776
            s->current_picture_ptr->frame_num= h->prev_frame_num;
3777
            execute_ref_pic_marking(h, NULL, 0);
3778
        }
3779

    
3780
        /* See if we have a decoded first field looking for a pair... */
3781
        if (s0->first_field) {
3782
            assert(s0->current_picture_ptr);
3783
            assert(s0->current_picture_ptr->data[0]);
3784
            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
3785

    
3786
            /* figure out if we have a complementary field pair */
3787
            if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
3788
                /*
3789
                 * Previous field is unmatched. Don't display it, but let it
3790
                 * remain for reference if marked as such.
3791
                 */
3792
                s0->current_picture_ptr = NULL;
3793
                s0->first_field = FIELD_PICTURE;
3794

    
3795
            } else {
3796
                if (h->nal_ref_idc &&
3797
                        s0->current_picture_ptr->reference &&
3798
                        s0->current_picture_ptr->frame_num != h->frame_num) {
3799
                    /*
3800
                     * This and previous field were reference, but had
3801
                     * different frame_nums. Consider this field first in
3802
                     * pair. Throw away previous field except for reference
3803
                     * purposes.
3804
                     */
3805
                    s0->first_field = 1;
3806
                    s0->current_picture_ptr = NULL;
3807

    
3808
                } else {
3809
                    /* Second field in complementary pair */
3810
                    s0->first_field = 0;
3811
                }
3812
            }
3813

    
3814
        } else {
3815
            /* Frame or first field in a potentially complementary pair */
3816
            assert(!s0->current_picture_ptr);
3817
            s0->first_field = FIELD_PICTURE;
3818
        }
3819

    
3820
        if((!FIELD_PICTURE || s0->first_field) && frame_start(h) < 0) {
3821
            s0->first_field = 0;
3822
            return -1;
3823
        }
3824
    }
3825
    if(h != h0)
3826
        clone_slice(h, h0);
3827

    
3828
    s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
3829

    
3830
    assert(s->mb_num == s->mb_width * s->mb_height);
3831
    if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
3832
       first_mb_in_slice                    >= s->mb_num){
3833
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3834
        return -1;
3835
    }
3836
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3837
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
3838
    if (s->picture_structure == PICT_BOTTOM_FIELD)
3839
        s->resync_mb_y = s->mb_y = s->mb_y + 1;
3840
    assert(s->mb_y < s->mb_height);
3841

    
3842
    if(s->picture_structure==PICT_FRAME){
3843
        h->curr_pic_num=   h->frame_num;
3844
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3845
    }else{
3846
        h->curr_pic_num= 2*h->frame_num + 1;
3847
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3848
    }
3849

    
3850
    if(h->nal_unit_type == NAL_IDR_SLICE){
3851
        get_ue_golomb(&s->gb); /* idr_pic_id */
3852
    }
3853

    
3854
    if(h->sps.poc_type==0){
3855
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3856

    
3857
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3858
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3859
        }
3860
    }
3861

    
3862
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3863
        h->delta_poc[0]= get_se_golomb(&s->gb);
3864

    
3865
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3866
            h->delta_poc[1]= get_se_golomb(&s->gb);
3867
    }
3868

    
3869
    init_poc(h);
3870

    
3871
    if(h->pps.redundant_pic_cnt_present){
3872
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3873
    }
3874

    
3875
    //set defaults, might be overridden a few lines later
3876
    h->ref_count[0]= h->pps.ref_count[0];
3877
    h->ref_count[1]= h->pps.ref_count[1];
3878

    
3879
    if(h->slice_type_nos != FF_I_TYPE){
3880
        if(h->slice_type_nos == FF_B_TYPE){
3881
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3882
        }
3883
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3884

    
3885
        if(num_ref_idx_active_override_flag){
3886
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3887
            if(h->slice_type_nos==FF_B_TYPE)
3888
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3889

    
3890
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
3891
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3892
                h->ref_count[0]= h->ref_count[1]= 1;
3893
                return -1;
3894
            }
3895
        }
3896
        if(h->slice_type_nos == FF_B_TYPE)
3897
            h->list_count= 2;
3898
        else
3899
            h->list_count= 1;
3900
    }else
3901
        h->list_count= 0;
3902

    
3903
    if(!default_ref_list_done){
3904
        fill_default_ref_list(h);
3905
    }
3906

    
3907
    if(h->slice_type_nos!=FF_I_TYPE && decode_ref_pic_list_reordering(h) < 0)
3908
        return -1;
3909

    
3910
    if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
3911
       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
3912
        pred_weight_table(h);
3913
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
3914
        implicit_weight_table(h);
3915
    else
3916
        h->use_weight = 0;
3917

    
3918
    if(h->nal_ref_idc)
3919
        decode_ref_pic_marking(h0, &s->gb);
3920

    
3921
    if(FRAME_MBAFF)
3922
        fill_mbaff_ref_list(h);
3923

    
3924
    if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
3925
        direct_dist_scale_factor(h);
3926
    direct_ref_list_init(h);
3927

    
3928
    if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
3929
        tmp = get_ue_golomb(&s->gb);
3930
        if(tmp > 2){
3931
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3932
            return -1;
3933
        }
3934
        h->cabac_init_idc= tmp;
3935
    }
3936

    
3937
    h->last_qscale_diff = 0;
3938
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3939
    if(tmp>51){
3940
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3941
        return -1;
3942
    }
3943
    s->qscale= tmp;
3944
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3945
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3946
    //FIXME qscale / qp ... stuff
3947
    if(h->slice_type == FF_SP_TYPE){
3948
        get_bits1(&s->gb); /* sp_for_switch_flag */
3949
    }
3950
    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
3951
        get_se_golomb(&s->gb); /* slice_qs_delta */
3952
    }
3953

    
3954
    h->deblocking_filter = 1;
3955
    h->slice_alpha_c0_offset = 0;
3956
    h->slice_beta_offset = 0;
3957
    if( h->pps.deblocking_filter_parameters_present ) {
3958
        tmp= get_ue_golomb(&s->gb);
3959
        if(tmp > 2){
3960
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3961
            return -1;
3962
        }
3963
        h->deblocking_filter= tmp;
3964
        if(h->deblocking_filter < 2)
3965
            h->deblocking_filter^= 1; // 1<->0
3966

    
3967
        if( h->deblocking_filter ) {
3968
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3969
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3970
        }
3971
    }
3972

    
3973
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
3974
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
3975
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
3976
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3977
        h->deblocking_filter= 0;
3978

    
3979
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
3980
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3981
            /* Cheat slightly for speed:
3982
               Do not bother to deblock across slices. */
3983
            h->deblocking_filter = 2;
3984
        } else {
3985
            h0->max_contexts = 1;
3986
            if(!h0->single_decode_warning) {
3987
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3988
                h0->single_decode_warning = 1;
3989
            }
3990
            if(h != h0)
3991
                return 1; // deblocking switched inside frame
3992
        }
3993
    }
3994

    
3995
#if 0 //FMO
3996
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3997
        slice_group_change_cycle= get_bits(&s->gb, ?);
3998
#endif
3999

    
4000
    h0->last_slice_type = slice_type;
4001
    h->slice_num = ++h0->current_slice;
4002

    
4003
    for(j=0; j<2; j++){
4004
        int *ref2frm= h->ref2frm[h->slice_num&15][j];
4005
        ref2frm[0]=
4006
        ref2frm[1]= -1;
4007
        for(i=0; i<16; i++)
4008
            ref2frm[i+2]= 4*h->ref_list[j][i].frame_num
4009
                          +(h->ref_list[j][i].reference&3);
4010
        ref2frm[18+0]=
4011
        ref2frm[18+1]= -1;
4012
        for(i=16; i<48; i++)
4013
            ref2frm[i+4]= 4*h->ref_list[j][i].frame_num
4014
                          +(h->ref_list[j][i].reference&3);
4015
    }
4016

    
4017
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
4018
    h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
4019

    
4020
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4021
        av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
4022
               h->slice_num,
4023
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
4024
               first_mb_in_slice,
4025
               av_get_pict_type_char(h->slice_type),
4026
               pps_id, h->frame_num,
4027
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4028
               h->ref_count[0], h->ref_count[1],
4029
               s->qscale,
4030
               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4031
               h->use_weight,
4032
               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
4033
               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
4034
               );
4035
    }
4036

    
4037
    return 0;
4038
}
4039

    
4040
/**
4041
 *
4042
 */
4043
static inline int get_level_prefix(GetBitContext *gb){
4044
    unsigned int buf;
4045
    int log;
4046

    
4047
    OPEN_READER(re, gb);
4048
    UPDATE_CACHE(re, gb);
4049
    buf=GET_CACHE(re, gb);
4050

    
4051
    log= 32 - av_log2(buf);
4052
#ifdef TRACE
4053
    print_bin(buf>>(32-log), log);
4054
    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__);
4055
#endif
4056

    
4057
    LAST_SKIP_BITS(re, gb, log);
4058
    CLOSE_READER(re, gb);
4059

    
4060
    return log-1;
4061
}
4062

    
4063
static inline int get_dct8x8_allowed(H264Context *h){
4064
    int i;
4065
    for(i=0; i<4; i++){
4066
        if(!IS_SUB_8X8(h->sub_mb_type[i])
4067
           || (!h->sps.direct_8x8_inference_flag && IS_DIRECT(h->sub_mb_type[i])))
4068
            return 0;
4069
    }
4070
    return 1;
4071
}
4072

    
4073
/**
4074
 * decodes a residual block.
4075
 * @param n block index
4076
 * @param scantable scantable
4077
 * @param max_coeff number of coefficients in the block
4078
 * @return <0 if an error occurred
4079
 */
4080
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
4081
    MpegEncContext * const s = &h->s;
4082
    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};
4083
    int level[16];
4084
    int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;
4085

    
4086
    //FIXME put trailing_onex into the context
4087

    
4088
    if(n == CHROMA_DC_BLOCK_INDEX){
4089
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4090
        total_coeff= coeff_token>>2;
4091
    }else{
4092
        if(n == LUMA_DC_BLOCK_INDEX){
4093
            total_coeff= pred_non_zero_count(h, 0);
4094
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4095
            total_coeff= coeff_token>>2;
4096
        }else{
4097
            total_coeff= pred_non_zero_count(h, n);
4098
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4099
            total_coeff= coeff_token>>2;
4100
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4101
        }
4102
    }
4103

    
4104
    //FIXME set last_non_zero?
4105

    
4106
    if(total_coeff==0)
4107
        return 0;
4108
    if(total_coeff > (unsigned)max_coeff) {
4109
        av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
4110
        return -1;
4111
    }
4112

    
4113
    trailing_ones= coeff_token&3;
4114
    tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
4115
    assert(total_coeff<=16);
4116

    
4117
    for(i=0; i<trailing_ones; i++){
4118
        level[i]= 1 - 2*get_bits1(gb);
4119
    }
4120

    
4121
    if(i<total_coeff) {
4122
        int level_code, mask;
4123
        int suffix_length = total_coeff > 10 && trailing_ones < 3;
4124
        int prefix= get_level_prefix(gb);
4125

    
4126
        //first coefficient has suffix_length equal to 0 or 1
4127
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4128
            if(suffix_length)
4129
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4130
            else
4131
                level_code= (prefix<<suffix_length); //part
4132
        }else if(prefix==14){
4133
            if(suffix_length)
4134
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4135
            else
4136
                level_code= prefix + get_bits(gb, 4); //part
4137
        }else{
4138
            level_code= (15<<suffix_length) + get_bits(gb, prefix-3); //part
4139
            if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
4140
            if(prefix>=16)
4141
                level_code += (1<<(prefix-3))-4096;
4142
        }
4143

    
4144
        if(trailing_ones < 3) level_code += 2;
4145

    
4146
        suffix_length = 1;
4147
        if(level_code > 5)
4148
            suffix_length++;
4149
        mask= -(level_code&1);
4150
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
4151
        i++;
4152

    
4153
        //remaining coefficients have suffix_length > 0
4154
        for(;i<total_coeff;i++) {
4155
            static const int suffix_limit[7] = {0,5,11,23,47,95,INT_MAX };
4156
            prefix = get_level_prefix(gb);
4157
            if(prefix<15){
4158
                level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
4159
            }else{
4160
                level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
4161
                if(prefix>=16)
4162
                    level_code += (1<<(prefix-3))-4096;
4163
            }
4164
            mask= -(level_code&1);
4165
            level[i]= (((2+level_code)>>1) ^ mask) - mask;
4166
            if(level_code > suffix_limit[suffix_length])
4167
                suffix_length++;
4168
        }
4169
    }
4170

    
4171
    if(total_coeff == max_coeff)
4172
        zeros_left=0;
4173
    else{
4174
        if(n == CHROMA_DC_BLOCK_INDEX)
4175
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4176
        else
4177
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4178
    }
4179

    
4180
    coeff_num = zeros_left + total_coeff - 1;
4181
    j = scantable[coeff_num];
4182
    if(n > 24){
4183
        block[j] = level[0];
4184
        for(i=1;i<total_coeff;i++) {
4185
            if(zeros_left <= 0)
4186
                run_before = 0;
4187
            else if(zeros_left < 7){
4188
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4189
            }else{
4190
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4191
            }
4192
            zeros_left -= run_before;
4193
            coeff_num -= 1 + run_before;
4194
            j= scantable[ coeff_num ];
4195

    
4196
            block[j]= level[i];
4197
        }
4198
    }else{
4199
        block[j] = (level[0] * qmul[j] + 32)>>6;
4200
        for(i=1;i<total_coeff;i++) {
4201
            if(zeros_left <= 0)
4202
                run_before = 0;
4203
            else if(zeros_left < 7){
4204
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4205
            }else{
4206
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4207
            }
4208
            zeros_left -= run_before;
4209
            coeff_num -= 1 + run_before;
4210
            j= scantable[ coeff_num ];
4211

    
4212
            block[j]= (level[i] * qmul[j] + 32)>>6;
4213
        }
4214
    }
4215

    
4216
    if(zeros_left<0){
4217
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4218
        return -1;
4219
    }
4220

    
4221
    return 0;
4222
}
4223

    
4224
static void predict_field_decoding_flag(H264Context *h){
4225
    MpegEncContext * const s = &h->s;
4226
    const int mb_xy= h->mb_xy;
4227
    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
4228
                ? s->current_picture.mb_type[mb_xy-1]
4229
                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
4230
                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
4231
                : 0;
4232
    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4233
}
4234

    
4235
/**
4236
 * decodes a P_SKIP or B_SKIP macroblock
4237
 */
4238
static void decode_mb_skip(H264Context *h){
4239
    MpegEncContext * const s = &h->s;
4240
    const int mb_xy= h->mb_xy;
4241
    int mb_type=0;
4242

    
4243
    memset(h->non_zero_count[mb_xy], 0, 16);
4244
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4245

    
4246
    if(MB_FIELD)
4247
        mb_type|= MB_TYPE_INTERLACED;
4248

    
4249
    if( h->slice_type_nos == FF_B_TYPE )
4250
    {
4251
        // just for fill_caches. pred_direct_motion will set the real mb_type
4252
        mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4253

    
4254
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4255
        pred_direct_motion(h, &mb_type);
4256
        mb_type|= MB_TYPE_SKIP;
4257
    }
4258
    else
4259
    {
4260
        int mx, my;
4261
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4262

    
4263
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4264
        pred_pskip_motion(h, &mx, &my);
4265
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4266
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4267
    }
4268

    
4269
    write_back_motion(h, mb_type);
4270
    s->current_picture.mb_type[mb_xy]= mb_type;
4271
    s->current_picture.qscale_table[mb_xy]= s->qscale;
4272
    h->slice_table[ mb_xy ]= h->slice_num;
4273
    h->prev_mb_skipped= 1;
4274
}
4275

    
4276
/**
4277
 * decodes a macroblock
4278
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4279
 */
4280
static int decode_mb_cavlc(H264Context *h){
4281
    MpegEncContext * const s = &h->s;
4282
    int mb_xy;
4283
    int partition_count;
4284
    unsigned int mb_type, cbp;
4285
    int dct8x8_allowed= h->pps.transform_8x8_mode;
4286

    
4287
    mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4288

    
4289
    s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?
4290

    
4291
    tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4292
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4293
                down the code */
4294
    if(h->slice_type_nos != FF_I_TYPE){
4295
        if(s->mb_skip_run==-1)
4296
            s->mb_skip_run= get_ue_golomb(&s->gb);
4297

    
4298
        if (s->mb_skip_run--) {
4299
            if(FRAME_MBAFF && (s->mb_y&1) == 0){
4300
                if(s->mb_skip_run==0)
4301
                    h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4302
                else
4303
                    predict_field_decoding_flag(h);
4304
            }
4305
            decode_mb_skip(h);
4306
            return 0;
4307
        }
4308
    }
4309
    if(FRAME_MBAFF){
4310
        if( (s->mb_y&1) == 0 )
4311
            h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4312
    }
4313

    
4314
    h->prev_mb_skipped= 0;
4315

    
4316
    mb_type= get_ue_golomb(&s->gb);
4317
    if(h->slice_type_nos == FF_B_TYPE){
4318
        if(mb_type < 23){
4319
            partition_count= b_mb_type_info[mb_type].partition_count;
4320
            mb_type=         b_mb_type_info[mb_type].type;
4321
        }else{
4322
            mb_type -= 23;
4323
            goto decode_intra_mb;
4324
        }
4325
    }else if(h->slice_type_nos == FF_P_TYPE){
4326
        if(mb_type < 5){
4327
            partition_count= p_mb_type_info[mb_type].partition_count;
4328
            mb_type=         p_mb_type_info[mb_type].type;
4329
        }else{
4330
            mb_type -= 5;
4331
            goto decode_intra_mb;
4332
        }
4333
    }else{
4334
       assert(h->slice_type_nos == FF_I_TYPE);
4335
        if(h->slice_type == FF_SI_TYPE && mb_type)
4336
            mb_type--;
4337
decode_intra_mb:
4338
        if(mb_type > 25){
4339
            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);
4340
            return -1;
4341
        }
4342
        partition_count=0;
4343
        cbp= i_mb_type_info[mb_type].cbp;
4344
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4345
        mb_type= i_mb_type_info[mb_type].type;
4346
    }
4347

    
4348
    if(MB_FIELD)
4349
        mb_type |= MB_TYPE_INTERLACED;
4350

    
4351
    h->slice_table[ mb_xy ]= h->slice_num;
4352

    
4353
    if(IS_INTRA_PCM(mb_type)){
4354
        unsigned int x;
4355

    
4356
        // We assume these blocks are very rare so we do not optimize it.
4357
        align_get_bits(&s->gb);
4358

    
4359
        // The pixels are stored in the same order as levels in h->mb array.
4360
        for(x=0; x < (CHROMA ? 384 : 256); x++){
4361
            ((uint8_t*)h->mb)[x]= get_bits(&s->gb, 8);
4362
        }
4363

    
4364
        // In deblocking, the quantizer is 0
4365
        s->current_picture.qscale_table[mb_xy]= 0;
4366
        // All coeffs are present
4367
        memset(h->non_zero_count[mb_xy], 16, 16);
4368

    
4369
        s->current_picture.mb_type[mb_xy]= mb_type;
4370
        return 0;
4371
    }
4372

    
4373
    if(MB_MBAFF){
4374
        h->ref_count[0] <<= 1;
4375
        h->ref_count[1] <<= 1;
4376
    }
4377

    
4378
    fill_caches(h, mb_type, 0);
4379

    
4380
    //mb_pred
4381
    if(IS_INTRA(mb_type)){
4382
        int pred_mode;
4383
//            init_top_left_availability(h);
4384
        if(IS_INTRA4x4(mb_type)){
4385
            int i;
4386
            int di = 1;
4387
            if(dct8x8_allowed && get_bits1(&s->gb)){
4388
                mb_type |= MB_TYPE_8x8DCT;
4389
                di = 4;
4390
            }
4391

    
4392
//                fill_intra4x4_pred_table(h);
4393
            for(i=0; i<16; i+=di){
4394
                int mode= pred_intra_mode(h, i);
4395

    
4396
                if(!get_bits1(&s->gb)){
4397
                    const int rem_mode= get_bits(&s->gb, 3);
4398
                    mode = rem_mode + (rem_mode >= mode);
4399
                }
4400

    
4401
                if(di==4)
4402
                    fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
4403
                else
4404
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4405
            }
4406
            write_back_intra_pred_mode(h);
4407
            if( check_intra4x4_pred_mode(h) < 0)
4408
                return -1;
4409
        }else{
4410
            h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4411
            if(h->intra16x16_pred_mode < 0)
4412
                return -1;
4413
        }
4414
        if(CHROMA){
4415
            pred_mode= check_intra_pred_mode(h, get_ue_golomb(&s->gb));
4416
            if(pred_mode < 0)
4417
                return -1;
4418
            h->chroma_pred_mode= pred_mode;
4419
        }
4420
    }else if(partition_count==4){
4421
        int i, j, sub_partition_count[4], list, ref[2][4];
4422

    
4423
        if(h->slice_type_nos == FF_B_TYPE){
4424
            for(i=0; i<4; i++){
4425
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4426
                if(h->sub_mb_type[i] >=13){
4427
                    av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
4428
                    return -1;
4429
                }
4430
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4431
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4432
            }
4433
            if(   IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4434
               || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
4435
                pred_direct_motion(h, &mb_type);
4436
                h->ref_cache[0][scan8[4]] =
4437
                h->ref_cache[1][scan8[4]] =
4438
                h->ref_cache[0][scan8[12]] =
4439
                h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
4440
            }
4441
        }else{
4442
            assert(h->slice_type_nos == FF_P_TYPE); //FIXME SP correct ?
4443
            for(i=0; i<4; i++){
4444
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4445
                if(h->sub_mb_type[i] >=4){
4446
                    av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
4447
                    return -1;
4448
                }
4449
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4450
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4451
            }
4452
        }
4453

    
4454
        for(list=0; list<h->list_count; list++){
4455
            int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4456
            for(i=0; i<4; i++){
4457
                if(IS_DIRECT(h->sub_mb_type[i])) continue;
4458
                if(IS_DIR(h->sub_mb_type[i], 0, list)){
4459
                    unsigned int tmp = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4460
                    if(tmp>=ref_count){
4461
                        av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", tmp);
4462
                        return -1;
4463
                    }
4464
                    ref[list][i]= tmp;
4465
                }else{
4466
                 //FIXME
4467
                    ref[list][i] = -1;
4468