Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 6026a096

History | View | Annotate | Download (315 KB)

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

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

    
28
#include "internal.h"
29
#include "dsputil.h"
30
#include "avcodec.h"
31
#include "mpegvideo.h"
32
#include "h264.h"
33
#include "h264data.h"
34
#include "h264_parser.h"
35
#include "golomb.h"
36
#include "mathops.h"
37
#include "rectangle.h"
38
#include "vdpau_internal.h"
39

    
40
#include "cabac.h"
41
#if ARCH_X86
42
#include "x86/h264_i386.h"
43
#endif
44

    
45
//#undef NDEBUG
46
#include <assert.h>
47

    
48
/**
49
 * Value of Picture.reference when Picture is not a reference picture, but
50
 * is held for delayed output.
51
 */
52
#define DELAYED_PIC_REF 4
53

    
54
static VLC coeff_token_vlc[4];
55
static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
56
static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
57

    
58
static VLC chroma_dc_coeff_token_vlc;
59
static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
60
static const int chroma_dc_coeff_token_vlc_table_size = 256;
61

    
62
static VLC total_zeros_vlc[15];
63
static VLC_TYPE total_zeros_vlc_tables[15][512][2];
64
static const int total_zeros_vlc_tables_size = 512;
65

    
66
static VLC chroma_dc_total_zeros_vlc[3];
67
static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
68
static const int chroma_dc_total_zeros_vlc_tables_size = 8;
69

    
70
static VLC run_vlc[6];
71
static VLC_TYPE run_vlc_tables[6][8][2];
72
static const int run_vlc_tables_size = 8;
73

    
74
static VLC run7_vlc;
75
static VLC_TYPE run7_vlc_table[96][2];
76
static const int run7_vlc_table_size = 96;
77

    
78
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
79
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
80
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);
81
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);
82
static Picture * remove_long(H264Context *h, int i, int ref_mask);
83

    
84
static av_always_inline uint32_t pack16to32(int a, int b){
85
#ifdef WORDS_BIGENDIAN
86
   return (b&0xFFFF) + (a<<16);
87
#else
88
   return (a&0xFFFF) + (b<<16);
89
#endif
90
}
91

    
92
static const uint8_t rem6[52]={
93
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,
94
};
95

    
96
static const uint8_t div6[52]={
97
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,
98
};
99

    
100
static const uint8_t left_block_options[4][8]={
101
    {0,1,2,3,7,10,8,11},
102
    {2,2,3,3,8,11,8,11},
103
    {0,0,1,1,7,10,7,10},
104
    {0,2,0,2,7,10,7,10}
105
};
106

    
107
#define LEVEL_TAB_BITS 8
108
static int8_t cavlc_level_tab[7][1<<LEVEL_TAB_BITS][2];
109

    
110
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
111
    MpegEncContext * const s = &h->s;
112
    const int mb_xy= h->mb_xy;
113
    int topleft_xy, top_xy, topright_xy, left_xy[2];
114
    int topleft_type, top_type, topright_type, left_type[2];
115
    const uint8_t * left_block;
116
    int topleft_partition= -1;
117
    int i;
118

    
119
    top_xy     = mb_xy  - (s->mb_stride << FIELD_PICTURE);
120

    
121
    //FIXME deblocking could skip the intra and nnz parts.
122
    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
123
        return;
124

    
125
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
126
     * stuff, I can't imagine that these complex rules are worth it. */
127

    
128
    topleft_xy = top_xy - 1;
129
    topright_xy= top_xy + 1;
130
    left_xy[1] = left_xy[0] = mb_xy-1;
131
    left_block = left_block_options[0];
132
    if(FRAME_MBAFF){
133
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
134
        const int top_pair_xy      = pair_xy     - s->mb_stride;
135
        const int topleft_pair_xy  = top_pair_xy - 1;
136
        const int topright_pair_xy = top_pair_xy + 1;
137
        const int topleft_mb_field_flag  = IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
138
        const int top_mb_field_flag      = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
139
        const int topright_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
140
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
141
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
142
        const int bottom = (s->mb_y & 1);
143
        tprintf(s->avctx, "fill_caches: curr_mb_field_flag:%d, left_mb_field_flag:%d, topleft_mb_field_flag:%d, top_mb_field_flag:%d, topright_mb_field_flag:%d\n", curr_mb_field_flag, left_mb_field_flag, topleft_mb_field_flag, top_mb_field_flag, topright_mb_field_flag);
144

    
145
        if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
146
            top_xy -= s->mb_stride;
147
        }
148
        if (curr_mb_field_flag && (bottom || topleft_mb_field_flag)){
149
            topleft_xy -= s->mb_stride;
150
        } else if(bottom && !curr_mb_field_flag && left_mb_field_flag) {
151
            topleft_xy += s->mb_stride;
152
            // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
153
            topleft_partition = 0;
154
        }
155
        if (curr_mb_field_flag && (bottom || topright_mb_field_flag)){
156
            topright_xy -= s->mb_stride;
157
        }
158
        if (left_mb_field_flag != curr_mb_field_flag) {
159
            left_xy[1] = left_xy[0] = pair_xy - 1;
160
            if (curr_mb_field_flag) {
161
                left_xy[1] += s->mb_stride;
162
                left_block = left_block_options[3];
163
            } else {
164
                left_block= left_block_options[2 - bottom];
165
            }
166
        }
167
    }
168

    
169
    h->top_mb_xy = top_xy;
170
    h->left_mb_xy[0] = left_xy[0];
171
    h->left_mb_xy[1] = left_xy[1];
172
    if(for_deblock){
173
        topleft_type = 0;
174
        topright_type = 0;
175
        top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
176
        left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
177
        left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
178

    
179
        if(MB_MBAFF && !IS_INTRA(mb_type)){
180
            int list;
181
            for(list=0; list<h->list_count; list++){
182
                //These values where changed for ease of performing MC, we need to change them back
183
                //FIXME maybe we can make MC and loop filter use the same values or prevent
184
                //the MC code from changing ref_cache and rather use a temporary array.
185
                if(USES_LIST(mb_type,list)){
186
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
187
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
188
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
189
                    ref += h->b8_stride;
190
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
191
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
192
                }
193
            }
194
        }
195
    }else{
196
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
197
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
198
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
199
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
200
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
201

    
202
    if(IS_INTRA(mb_type)){
203
        int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
204
        h->topleft_samples_available=
205
        h->top_samples_available=
206
        h->left_samples_available= 0xFFFF;
207
        h->topright_samples_available= 0xEEEA;
208

    
209
        if(!(top_type & type_mask)){
210
            h->topleft_samples_available= 0xB3FF;
211
            h->top_samples_available= 0x33FF;
212
            h->topright_samples_available= 0x26EA;
213
        }
214
        if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
215
            if(IS_INTERLACED(mb_type)){
216
                if(!(left_type[0] & type_mask)){
217
                    h->topleft_samples_available&= 0xDFFF;
218
                    h->left_samples_available&= 0x5FFF;
219
                }
220
                if(!(left_type[1] & type_mask)){
221
                    h->topleft_samples_available&= 0xFF5F;
222
                    h->left_samples_available&= 0xFF5F;
223
                }
224
            }else{
225
                int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
226
                                ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
227
                assert(left_xy[0] == left_xy[1]);
228
                if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
229
                    h->topleft_samples_available&= 0xDF5F;
230
                    h->left_samples_available&= 0x5F5F;
231
                }
232
            }
233
        }else{
234
            if(!(left_type[0] & type_mask)){
235
                h->topleft_samples_available&= 0xDF5F;
236
                h->left_samples_available&= 0x5F5F;
237
            }
238
        }
239

    
240
        if(!(topleft_type & type_mask))
241
            h->topleft_samples_available&= 0x7FFF;
242

    
243
        if(!(topright_type & type_mask))
244
            h->topright_samples_available&= 0xFBFF;
245

    
246
        if(IS_INTRA4x4(mb_type)){
247
            if(IS_INTRA4x4(top_type)){
248
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
249
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
250
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
251
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
252
            }else{
253
                int pred;
254
                if(!(top_type & type_mask))
255
                    pred= -1;
256
                else{
257
                    pred= 2;
258
                }
259
                h->intra4x4_pred_mode_cache[4+8*0]=
260
                h->intra4x4_pred_mode_cache[5+8*0]=
261
                h->intra4x4_pred_mode_cache[6+8*0]=
262
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
263
            }
264
            for(i=0; i<2; i++){
265
                if(IS_INTRA4x4(left_type[i])){
266
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
267
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
268
                }else{
269
                    int pred;
270
                    if(!(left_type[i] & type_mask))
271
                        pred= -1;
272
                    else{
273
                        pred= 2;
274
                    }
275
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
276
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
277
                }
278
            }
279
        }
280
    }
281
    }
282

    
283

    
284
/*
285
0 . T T. T T T T
286
1 L . .L . . . .
287
2 L . .L . . . .
288
3 . T TL . . . .
289
4 L . .L . . . .
290
5 L . .. . . . .
291
*/
292
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
293
    if(top_type){
294
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
295
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
296
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
297
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
298

    
299
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
300
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
301

    
302
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
303
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
304

    
305
    }else{
306
        h->non_zero_count_cache[4+8*0]=
307
        h->non_zero_count_cache[5+8*0]=
308
        h->non_zero_count_cache[6+8*0]=
309
        h->non_zero_count_cache[7+8*0]=
310

    
311
        h->non_zero_count_cache[1+8*0]=
312
        h->non_zero_count_cache[2+8*0]=
313

    
314
        h->non_zero_count_cache[1+8*3]=
315
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
316

    
317
    }
318

    
319
    for (i=0; i<2; i++) {
320
        if(left_type[i]){
321
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
322
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
323
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
324
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
325
        }else{
326
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
327
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
328
            h->non_zero_count_cache[0+8*1 +   8*i]=
329
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
330
        }
331
    }
332

    
333
    if( h->pps.cabac ) {
334
        // top_cbp
335
        if(top_type) {
336
            h->top_cbp = h->cbp_table[top_xy];
337
        } else if(IS_INTRA(mb_type)) {
338
            h->top_cbp = 0x1C0;
339
        } else {
340
            h->top_cbp = 0;
341
        }
342
        // left_cbp
343
        if (left_type[0]) {
344
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
345
        } else if(IS_INTRA(mb_type)) {
346
            h->left_cbp = 0x1C0;
347
        } else {
348
            h->left_cbp = 0;
349
        }
350
        if (left_type[0]) {
351
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
352
        }
353
        if (left_type[1]) {
354
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
355
        }
356
    }
357

    
358
#if 1
359
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
360
        int list;
361
        for(list=0; list<h->list_count; list++){
362
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
363
                /*if(!h->mv_cache_clean[list]){
364
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
365
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
366
                    h->mv_cache_clean[list]= 1;
367
                }*/
368
                continue;
369
            }
370
            h->mv_cache_clean[list]= 0;
371

    
372
            if(USES_LIST(top_type, list)){
373
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
374
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
375
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
376
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
377
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
378
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
379
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
380
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
381
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
382
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
383
            }else{
384
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
385
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
386
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
387
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
388
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
389
            }
390

    
391
            for(i=0; i<2; i++){
392
                int cache_idx = scan8[0] - 1 + i*2*8;
393
                if(USES_LIST(left_type[i], list)){
394
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
395
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
396
                    *(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]];
397
                    *(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]];
398
                    h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
399
                    h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
400
                }else{
401
                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
402
                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
403
                    h->ref_cache[list][cache_idx  ]=
404
                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
405
                }
406
            }
407

    
408
            if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
409
                continue;
410

    
411
            if(USES_LIST(topleft_type, list)){
412
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
413
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
414
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
415
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
416
            }else{
417
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
418
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
419
            }
420

    
421
            if(USES_LIST(topright_type, list)){
422
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
423
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
424
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
425
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
426
            }else{
427
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
428
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
429
            }
430

    
431
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
432
                continue;
433

    
434
            h->ref_cache[list][scan8[5 ]+1] =
435
            h->ref_cache[list][scan8[7 ]+1] =
436
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
437
            h->ref_cache[list][scan8[4 ]] =
438
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
439
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
440
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
441
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
442
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
443
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
444

    
445
            if( h->pps.cabac ) {
446
                /* XXX beurk, Load mvd */
447
                if(USES_LIST(top_type, list)){
448
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
449
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
450
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
451
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
452
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
453
                }else{
454
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
455
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
456
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
457
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
458
                }
459
                if(USES_LIST(left_type[0], list)){
460
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
461
                    *(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]];
462
                    *(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]];
463
                }else{
464
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
465
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
466
                }
467
                if(USES_LIST(left_type[1], list)){
468
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
469
                    *(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]];
470
                    *(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]];
471
                }else{
472
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
473
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
474
                }
475
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
476
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
477
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
478
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
479
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
480

    
481
                if(h->slice_type_nos == FF_B_TYPE){
482
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
483

    
484
                    if(IS_DIRECT(top_type)){
485
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
486
                    }else if(IS_8X8(top_type)){
487
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
488
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
489
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
490
                    }else{
491
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
492
                    }
493

    
494
                    if(IS_DIRECT(left_type[0]))
495
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
496
                    else if(IS_8X8(left_type[0]))
497
                        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)];
498
                    else
499
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
500

    
501
                    if(IS_DIRECT(left_type[1]))
502
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
503
                    else if(IS_8X8(left_type[1]))
504
                        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)];
505
                    else
506
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
507
                }
508
            }
509

    
510
            if(FRAME_MBAFF){
511
#define MAP_MVS\
512
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
513
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
514
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
515
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
516
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
517
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
518
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
519
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
520
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
521
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
522
                if(MB_FIELD){
523
#define MAP_F2F(idx, mb_type)\
524
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
525
                        h->ref_cache[list][idx] <<= 1;\
526
                        h->mv_cache[list][idx][1] /= 2;\
527
                        h->mvd_cache[list][idx][1] /= 2;\
528
                    }
529
                    MAP_MVS
530
#undef MAP_F2F
531
                }else{
532
#define MAP_F2F(idx, mb_type)\
533
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
534
                        h->ref_cache[list][idx] >>= 1;\
535
                        h->mv_cache[list][idx][1] <<= 1;\
536
                        h->mvd_cache[list][idx][1] <<= 1;\
537
                    }
538
                    MAP_MVS
539
#undef MAP_F2F
540
                }
541
            }
542
        }
543
    }
544
#endif
545

    
546
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
547
}
548

    
549
static inline void write_back_intra_pred_mode(H264Context *h){
550
    const int mb_xy= h->mb_xy;
551

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

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

    
570
    if(!(h->top_samples_available&0x8000)){
571
        for(i=0; i<4; i++){
572
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
573
            if(status<0){
574
                av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
575
                return -1;
576
            } else if(status){
577
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
578
            }
579
        }
580
    }
581

    
582
    if((h->left_samples_available&0x8888)!=0x8888){
583
        static const int mask[4]={0x8000,0x2000,0x80,0x20};
584
        for(i=0; i<4; i++){
585
            if(!(h->left_samples_available&mask[i])){
586
                int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
587
                if(status<0){
588
                    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);
589
                    return -1;
590
                } else if(status){
591
                    h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
592
                }
593
            }
594
        }
595
    }
596

    
597
    return 0;
598
} //FIXME cleanup like next
599

    
600
/**
601
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
602
 */
603
static inline int check_intra_pred_mode(H264Context *h, int mode){
604
    MpegEncContext * const s = &h->s;
605
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
606
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
607

    
608
    if(mode > 6U) {
609
        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);
610
        return -1;
611
    }
612

    
613
    if(!(h->top_samples_available&0x8000)){
614
        mode= top[ mode ];
615
        if(mode<0){
616
            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);
617
            return -1;
618
        }
619
    }
620

    
621
    if((h->left_samples_available&0x8080) != 0x8080){
622
        mode= left[ mode ];
623
        if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
624
            mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
625
        }
626
        if(mode<0){
627
            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);
628
            return -1;
629
        }
630
    }
631

    
632
    return mode;
633
}
634

    
635
/**
636
 * gets the predicted intra4x4 prediction mode.
637
 */
638
static inline int pred_intra_mode(H264Context *h, int n){
639
    const int index8= scan8[n];
640
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
641
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
642
    const int min= FFMIN(left, top);
643

    
644
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
645

    
646
    if(min<0) return DC_PRED;
647
    else      return min;
648
}
649

    
650
static inline void write_back_non_zero_count(H264Context *h){
651
    const int mb_xy= h->mb_xy;
652

    
653
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
654
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
655
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
656
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
657
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
658
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
659
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
660

    
661
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
662
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
663
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
664

    
665
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
666
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
667
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
668
}
669

    
670
/**
671
 * gets the predicted number of non-zero coefficients.
672
 * @param n block index
673
 */
674
static inline int pred_non_zero_count(H264Context *h, int n){
675
    const int index8= scan8[n];
676
    const int left= h->non_zero_count_cache[index8 - 1];
677
    const int top = h->non_zero_count_cache[index8 - 8];
678
    int i= left + top;
679

    
680
    if(i<64) i= (i+1)>>1;
681

    
682
    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
683

    
684
    return i&31;
685
}
686

    
687
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
688
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
689
    MpegEncContext *s = &h->s;
690

    
691
    /* there is no consistent mapping of mvs to neighboring locations that will
692
     * make mbaff happy, so we can't move all this logic to fill_caches */
693
    if(FRAME_MBAFF){
694
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
695
        const int16_t *mv;
696
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
697
        *C = h->mv_cache[list][scan8[0]-2];
698

    
699
        if(!MB_FIELD
700
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
701
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
702
            if(IS_INTERLACED(mb_types[topright_xy])){
703
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
704
                const int x4 = X4, y4 = Y4;\
705
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
706
                if(!USES_LIST(mb_type,list))\
707
                    return LIST_NOT_USED;\
708
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
709
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
710
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
711
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
712

    
713
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
714
            }
715
        }
716
        if(topright_ref == PART_NOT_AVAILABLE
717
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
718
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
719
            if(!MB_FIELD
720
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
721
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
722
            }
723
            if(MB_FIELD
724
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
725
               && i >= scan8[0]+8){
726
                // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
727
                SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
728
            }
729
        }
730
#undef SET_DIAG_MV
731
    }
732

    
733
    if(topright_ref != PART_NOT_AVAILABLE){
734
        *C= h->mv_cache[list][ i - 8 + part_width ];
735
        return topright_ref;
736
    }else{
737
        tprintf(s->avctx, "topright MV not available\n");
738

    
739
        *C= h->mv_cache[list][ i - 8 - 1 ];
740
        return h->ref_cache[list][ i - 8 - 1 ];
741
    }
742
}
743

    
744
/**
745
 * gets the predicted MV.
746
 * @param n the block index
747
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
748
 * @param mx the x component of the predicted motion vector
749
 * @param my the y component of the predicted motion vector
750
 */
751
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
752
    const int index8= scan8[n];
753
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
754
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
755
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
756
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
757
    const int16_t * C;
758
    int diagonal_ref, match_count;
759

    
760
    assert(part_width==1 || part_width==2 || part_width==4);
761

    
762
/* mv_cache
763
  B . . A T T T T
764
  U . . L . . , .
765
  U . . L . . . .
766
  U . . L . . , .
767
  . . . L . . . .
768
*/
769

    
770
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
771
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
772
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
773
    if(match_count > 1){ //most common
774
        *mx= mid_pred(A[0], B[0], C[0]);
775
        *my= mid_pred(A[1], B[1], C[1]);
776
    }else if(match_count==1){
777
        if(left_ref==ref){
778
            *mx= A[0];
779
            *my= A[1];
780
        }else if(top_ref==ref){
781
            *mx= B[0];
782
            *my= B[1];
783
        }else{
784
            *mx= C[0];
785
            *my= C[1];
786
        }
787
    }else{
788
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
789
            *mx= A[0];
790
            *my= A[1];
791
        }else{
792
            *mx= mid_pred(A[0], B[0], C[0]);
793
            *my= mid_pred(A[1], B[1], C[1]);
794
        }
795
    }
796

    
797
    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);
798
}
799

    
800
/**
801
 * gets the directionally predicted 16x8 MV.
802
 * @param n the block index
803
 * @param mx the x component of the predicted motion vector
804
 * @param my the y component of the predicted motion vector
805
 */
806
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
807
    if(n==0){
808
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
809
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
810

    
811
        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);
812

    
813
        if(top_ref == ref){
814
            *mx= B[0];
815
            *my= B[1];
816
            return;
817
        }
818
    }else{
819
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
820
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
821

    
822
        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);
823

    
824
        if(left_ref == ref){
825
            *mx= A[0];
826
            *my= A[1];
827
            return;
828
        }
829
    }
830

    
831
    //RARE
832
    pred_motion(h, n, 4, list, ref, mx, my);
833
}
834

    
835
/**
836
 * gets the directionally predicted 8x16 MV.
837
 * @param n the block index
838
 * @param mx the x component of the predicted motion vector
839
 * @param my the y component of the predicted motion vector
840
 */
841
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
842
    if(n==0){
843
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
844
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
845

    
846
        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);
847

    
848
        if(left_ref == ref){
849
            *mx= A[0];
850
            *my= A[1];
851
            return;
852
        }
853
    }else{
854
        const int16_t * C;
855
        int diagonal_ref;
856

    
857
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
858

    
859
        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);
860

    
861
        if(diagonal_ref == ref){
862
            *mx= C[0];
863
            *my= C[1];
864
            return;
865
        }
866
    }
867

    
868
    //RARE
869
    pred_motion(h, n, 2, list, ref, mx, my);
870
}
871

    
872
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
873
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
874
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
875

    
876
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
877

    
878
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
879
       || !( top_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ])
880
       || !(left_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ])){
881

    
882
        *mx = *my = 0;
883
        return;
884
    }
885

    
886
    pred_motion(h, 0, 4, 0, 0, mx, my);
887

    
888
    return;
889
}
890

    
891
static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
892
    int poc0 = h->ref_list[0][i].poc;
893
    int td = av_clip(poc1 - poc0, -128, 127);
894
    if(td == 0 || h->ref_list[0][i].long_ref){
895
        return 256;
896
    }else{
897
        int tb = av_clip(poc - poc0, -128, 127);
898
        int tx = (16384 + (FFABS(td) >> 1)) / td;
899
        return av_clip((tb*tx + 32) >> 6, -1024, 1023);
900
    }
901
}
902

    
903
static inline void direct_dist_scale_factor(H264Context * const h){
904
    MpegEncContext * const s = &h->s;
905
    const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
906
    const int poc1 = h->ref_list[1][0].poc;
907
    int i, field;
908
    for(field=0; field<2; field++){
909
        const int poc  = h->s.current_picture_ptr->field_poc[field];
910
        const int poc1 = h->ref_list[1][0].field_poc[field];
911
        for(i=0; i < 2*h->ref_count[0]; i++)
912
            h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
913
    }
914

    
915
    for(i=0; i<h->ref_count[0]; i++){
916
        h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
917
    }
918
}
919

    
920
static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
921
    MpegEncContext * const s = &h->s;
922
    Picture * const ref1 = &h->ref_list[1][0];
923
    int j, old_ref, rfield;
924
    int start= mbafi ? 16                      : 0;
925
    int end  = mbafi ? 16+2*h->ref_count[list] : h->ref_count[list];
926
    int interl= mbafi || s->picture_structure != PICT_FRAME;
927

    
928
    /* bogus; fills in for missing frames */
929
    memset(map[list], 0, sizeof(map[list]));
930

    
931
    for(rfield=0; rfield<2; rfield++){
932
        for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
933
            int poc = ref1->ref_poc[colfield][list][old_ref];
934

    
935
            if     (!interl)
936
                poc |= 3;
937
            else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
938
                poc= (poc&~3) + rfield + 1;
939

    
940
            for(j=start; j<end; j++){
941
                if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
942
                    int cur_ref= mbafi ? (j-16)^field : j;
943
                    map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
944
                    if(rfield == field)
945
                        map[list][old_ref] = cur_ref;
946
                    break;
947
                }
948
            }
949
        }
950
    }
951
}
952

    
953
static inline void direct_ref_list_init(H264Context * const h){
954
    MpegEncContext * const s = &h->s;
955
    Picture * const ref1 = &h->ref_list[1][0];
956
    Picture * const cur = s->current_picture_ptr;
957
    int list, j, field;
958
    int sidx= (s->picture_structure&1)^1;
959
    int ref1sidx= (ref1->reference&1)^1;
960

    
961
    for(list=0; list<2; list++){
962
        cur->ref_count[sidx][list] = h->ref_count[list];
963
        for(j=0; j<h->ref_count[list]; j++)
964
            cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
965
    }
966

    
967
    if(s->picture_structure == PICT_FRAME){
968
        memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
969
        memcpy(cur->ref_poc  [1], cur->ref_poc  [0], sizeof(cur->ref_poc  [0]));
970
    }
971

    
972
    cur->mbaff= FRAME_MBAFF;
973

    
974
    if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
975
        return;
976

    
977
    for(list=0; list<2; list++){
978
        fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
979
        for(field=0; field<2; field++)
980
            fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
981
    }
982
}
983

    
984
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
985
    MpegEncContext * const s = &h->s;
986
    int b8_stride = h->b8_stride;
987
    int b4_stride = h->b_stride;
988
    int mb_xy = h->mb_xy;
989
    int mb_type_col[2];
990
    const int16_t (*l1mv0)[2], (*l1mv1)[2];
991
    const int8_t *l1ref0, *l1ref1;
992
    const int is_b8x8 = IS_8X8(*mb_type);
993
    unsigned int sub_mb_type;
994
    int i8, i4;
995

    
996
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
997

    
998
    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
999
        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
1000
            int cur_poc = s->current_picture_ptr->poc;
1001
            int *col_poc = h->ref_list[1]->field_poc;
1002
            int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
1003
            mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
1004
            b8_stride = 0;
1005
        }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity
1006
            int fieldoff= 2*(h->ref_list[1][0].reference)-3;
1007
            mb_xy += s->mb_stride*fieldoff;
1008
        }
1009
        goto single_col;
1010
    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
1011
        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
1012
            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
1013
            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
1014
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
1015
            b8_stride *= 3;
1016
            b4_stride *= 6;
1017
            //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag
1018
            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
1019
                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
1020
                && !is_b8x8){
1021
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1022
                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
1023
            }else{
1024
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1025
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1026
            }
1027
        }else{                                           //     AFR/FR    -> AFR/FR
1028
single_col:
1029
            mb_type_col[0] =
1030
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
1031
            if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
1032
                /* FIXME save sub mb types from previous frames (or derive from MVs)
1033
                * so we know exactly what block size to use */
1034
                sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1035
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1036
            }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
1037
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1038
                *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1039
            }else{
1040
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1041
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1042
            }
1043
        }
1044
    }
1045

    
1046
    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
1047
    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
1048
    l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
1049
    l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
1050
    if(!b8_stride){
1051
        if(s->mb_y&1){
1052
            l1ref0 += h->b8_stride;
1053
            l1ref1 += h->b8_stride;
1054
            l1mv0  +=  2*b4_stride;
1055
            l1mv1  +=  2*b4_stride;
1056
        }
1057
    }
1058

    
1059
    if(h->direct_spatial_mv_pred){
1060
        int ref[2];
1061
        int mv[2][2];
1062
        int list;
1063

    
1064
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1065

    
1066
        /* ref = min(neighbors) */
1067
        for(list=0; list<2; list++){
1068
            int refa = h->ref_cache[list][scan8[0] - 1];
1069
            int refb = h->ref_cache[list][scan8[0] - 8];
1070
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1071
            if(refc == PART_NOT_AVAILABLE)
1072
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1073
            ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
1074
            if(ref[list] < 0)
1075
                ref[list] = -1;
1076
        }
1077

    
1078
        if(ref[0] < 0 && ref[1] < 0){
1079
            ref[0] = ref[1] = 0;
1080
            mv[0][0] = mv[0][1] =
1081
            mv[1][0] = mv[1][1] = 0;
1082
        }else{
1083
            for(list=0; list<2; list++){
1084
                if(ref[list] >= 0)
1085
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1086
                else
1087
                    mv[list][0] = mv[list][1] = 0;
1088
            }
1089
        }
1090

    
1091
        if(ref[1] < 0){
1092
            if(!is_b8x8)
1093
                *mb_type &= ~MB_TYPE_L1;
1094
            sub_mb_type &= ~MB_TYPE_L1;
1095
        }else if(ref[0] < 0){
1096
            if(!is_b8x8)
1097
                *mb_type &= ~MB_TYPE_L0;
1098
            sub_mb_type &= ~MB_TYPE_L0;
1099
        }
1100

    
1101
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1102
            for(i8=0; i8<4; i8++){
1103
                int x8 = i8&1;
1104
                int y8 = i8>>1;
1105
                int xy8 = x8+y8*b8_stride;
1106
                int xy4 = 3*x8+y8*b4_stride;
1107
                int a=0, b=0;
1108

    
1109
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1110
                    continue;
1111
                h->sub_mb_type[i8] = sub_mb_type;
1112

    
1113
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1114
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1115
                if(!IS_INTRA(mb_type_col[y8])
1116
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1117
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1118
                    if(ref[0] > 0)
1119
                        a= pack16to32(mv[0][0],mv[0][1]);
1120
                    if(ref[1] > 0)
1121
                        b= pack16to32(mv[1][0],mv[1][1]);
1122
                }else{
1123
                    a= pack16to32(mv[0][0],mv[0][1]);
1124
                    b= pack16to32(mv[1][0],mv[1][1]);
1125
                }
1126
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1127
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1128
            }
1129
        }else if(IS_16X16(*mb_type)){
1130
            int a=0, b=0;
1131

    
1132
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1133
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1134
            if(!IS_INTRA(mb_type_col[0])
1135
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1136
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1137
                       && (h->x264_build>33 || !h->x264_build)))){
1138
                if(ref[0] > 0)
1139
                    a= pack16to32(mv[0][0],mv[0][1]);
1140
                if(ref[1] > 0)
1141
                    b= pack16to32(mv[1][0],mv[1][1]);
1142
            }else{
1143
                a= pack16to32(mv[0][0],mv[0][1]);
1144
                b= pack16to32(mv[1][0],mv[1][1]);
1145
            }
1146
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1147
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1148
        }else{
1149
            for(i8=0; i8<4; i8++){
1150
                const int x8 = i8&1;
1151
                const int y8 = i8>>1;
1152

    
1153
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1154
                    continue;
1155
                h->sub_mb_type[i8] = sub_mb_type;
1156

    
1157
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1158
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1159
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1160
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1161

    
1162
                /* col_zero_flag */
1163
                if(!IS_INTRA(mb_type_col[0]) && (   l1ref0[x8 + y8*b8_stride] == 0
1164
                                              || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
1165
                                                  && (h->x264_build>33 || !h->x264_build)))){
1166
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
1167
                    if(IS_SUB_8X8(sub_mb_type)){
1168
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1169
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1170
                            if(ref[0] == 0)
1171
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1172
                            if(ref[1] == 0)
1173
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1174
                        }
1175
                    }else
1176
                    for(i4=0; i4<4; i4++){
1177
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1178
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1179
                            if(ref[0] == 0)
1180
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1181
                            if(ref[1] == 0)
1182
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1183
                        }
1184
                    }
1185
                }
1186
            }
1187
        }
1188
    }else{ /* direct temporal mv pred */
1189
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1190
        const int *dist_scale_factor = h->dist_scale_factor;
1191
        int ref_offset= 0;
1192

    
1193
        if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
1194
            map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
1195
            map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
1196
            dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];
1197
        }
1198
        if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
1199
            ref_offset += 16;
1200

    
1201
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1202
            /* FIXME assumes direct_8x8_inference == 1 */
1203
            int y_shift  = 2*!IS_INTERLACED(*mb_type);
1204

    
1205
            for(i8=0; i8<4; i8++){
1206
                const int x8 = i8&1;
1207
                const int y8 = i8>>1;
1208
                int ref0, scale;
1209
                const int16_t (*l1mv)[2]= l1mv0;
1210

    
1211
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1212
                    continue;
1213
                h->sub_mb_type[i8] = sub_mb_type;
1214

    
1215
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1216
                if(IS_INTRA(mb_type_col[y8])){
1217
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1218
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1219
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1220
                    continue;
1221
                }
1222

    
1223
                ref0 = l1ref0[x8 + y8*b8_stride];
1224
                if(ref0 >= 0)
1225
                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
1226
                else{
1227
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1228
                    l1mv= l1mv1;
1229
                }
1230
                scale = dist_scale_factor[ref0];
1231
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1232

    
1233
                {
1234
                    const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
1235
                    int my_col = (mv_col[1]<<y_shift)/2;
1236
                    int mx = (scale * mv_col[0] + 128) >> 8;
1237
                    int my = (scale * my_col + 128) >> 8;
1238
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1239
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1240
                }
1241
            }
1242
            return;
1243
        }
1244

    
1245
        /* one-to-one mv scaling */
1246

    
1247
        if(IS_16X16(*mb_type)){
1248
            int ref, mv0, mv1;
1249

    
1250
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1251
            if(IS_INTRA(mb_type_col[0])){
1252
                ref=mv0=mv1=0;
1253
            }else{
1254
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
1255
                                                : map_col_to_list0[1][l1ref1[0] + ref_offset];
1256
                const int scale = dist_scale_factor[ref0];
1257
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1258
                int mv_l0[2];
1259
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1260
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1261
                ref= ref0;
1262
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
1263
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1264
            }
1265
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1266
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1267
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1268
        }else{
1269
            for(i8=0; i8<4; i8++){
1270
                const int x8 = i8&1;
1271
                const int y8 = i8>>1;
1272
                int ref0, scale;
1273
                const int16_t (*l1mv)[2]= l1mv0;
1274

    
1275
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1276
                    continue;
1277
                h->sub_mb_type[i8] = sub_mb_type;
1278
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1279
                if(IS_INTRA(mb_type_col[0])){
1280
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1281
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1282
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1283
                    continue;
1284
                }
1285

    
1286
                ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset;
1287
                if(ref0 >= 0)
1288
                    ref0 = map_col_to_list0[0][ref0];
1289
                else{
1290
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1291
                    l1mv= l1mv1;
1292
                }
1293
                scale = dist_scale_factor[ref0];
1294

    
1295
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1296
                if(IS_SUB_8X8(sub_mb_type)){
1297
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1298
                    int mx = (scale * mv_col[0] + 128) >> 8;
1299
                    int my = (scale * mv_col[1] + 128) >> 8;
1300
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1301
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1302
                }else
1303
                for(i4=0; i4<4; i4++){
1304
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1305
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1306
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1307
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1308
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1309
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1310
                }
1311
            }
1312
        }
1313
    }
1314
}
1315

    
1316
static inline void write_back_motion(H264Context *h, int mb_type){
1317
    MpegEncContext * const s = &h->s;
1318
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1319
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1320
    int list;
1321

    
1322
    if(!USES_LIST(mb_type, 0))
1323
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1324

    
1325
    for(list=0; list<h->list_count; list++){
1326
        int y;
1327
        if(!USES_LIST(mb_type, list))
1328
            continue;
1329

    
1330
        for(y=0; y<4; y++){
1331
            *(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];
1332
            *(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];
1333
        }
1334
        if( h->pps.cabac ) {
1335
            if(IS_SKIP(mb_type))
1336
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1337
            else
1338
            for(y=0; y<4; y++){
1339
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1340
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1341
            }
1342
        }
1343

    
1344
        {
1345
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1346
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1347
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1348
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1349
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1350
        }
1351
    }
1352

    
1353
    if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
1354
        if(IS_8X8(mb_type)){
1355
            uint8_t *direct_table = &h->direct_table[b8_xy];
1356
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1357
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1358
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1359
        }
1360
    }
1361
}
1362

    
1363
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1364
    int i, si, di;
1365
    uint8_t *dst;
1366
    int bufidx;
1367

    
1368
//    src[0]&0x80;                //forbidden bit
1369
    h->nal_ref_idc= src[0]>>5;
1370
    h->nal_unit_type= src[0]&0x1F;
1371

    
1372
    src++; length--;
1373
#if 0
1374
    for(i=0; i<length; i++)
1375
        printf("%2X ", src[i]);
1376
#endif
1377

    
1378
#if HAVE_FAST_UNALIGNED
1379
# if HAVE_FAST_64BIT
1380
#   define RS 7
1381
    for(i=0; i+1<length; i+=9){
1382
        if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
1383
# else
1384
#   define RS 3
1385
    for(i=0; i+1<length; i+=5){
1386
        if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
1387
# endif
1388
            continue;
1389
        if(i>0 && !src[i]) i--;
1390
        while(src[i]) i++;
1391
#else
1392
#   define RS 0
1393
    for(i=0; i+1<length; i+=2){
1394
        if(src[i]) continue;
1395
        if(i>0 && src[i-1]==0) i--;
1396
#endif
1397
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1398
            if(src[i+2]!=3){
1399
                /* startcode, so we must be past the end */
1400
                length=i;
1401
            }
1402
            break;
1403
        }
1404
        i-= RS;
1405
    }
1406

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

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

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

    
1421
//printf("decoding esc\n");
1422
    memcpy(dst, src, i);
1423
    si=di=i;
1424
    while(si+2<length){
1425
        //remove escapes (very rare 1:2^22)
1426
        if(src[si+2]>3){
1427
            dst[di++]= src[si++];
1428
            dst[di++]= src[si++];
1429
        }else if(src[si]==0 && src[si+1]==0){
1430
            if(src[si+2]==3){ //escape
1431
                dst[di++]= 0;
1432
                dst[di++]= 0;
1433
                si+=3;
1434
                continue;
1435
            }else //next start code
1436
                goto nsc;
1437
        }
1438

    
1439
        dst[di++]= src[si++];
1440
    }
1441
    while(si<length)
1442
        dst[di++]= src[si++];
1443
nsc:
1444

    
1445
    memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
1446

    
1447
    *dst_length= di;
1448
    *consumed= si + 1;//+1 for the header
1449
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1450
    return dst;
1451
}
1452

    
1453
int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1454
    int v= *src;
1455
    int r;
1456

    
1457
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1458

    
1459
    for(r=1; r<9; r++){
1460
        if(v&1) return r;
1461
        v>>=1;
1462
    }
1463
    return 0;
1464
}
1465

    
1466
/**
1467
 * IDCT transforms the 16 dc values and dequantizes them.
1468
 * @param qp quantization parameter
1469
 */
1470
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1471
#define stride 16
1472
    int i;
1473
    int temp[16]; //FIXME check if this is a good idea
1474
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1475
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1476

    
1477
//memset(block, 64, 2*256);
1478
//return;
1479
    for(i=0; i<4; i++){
1480
        const int offset= y_offset[i];
1481
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1482
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1483
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1484
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1485

    
1486
        temp[4*i+0]= z0+z3;
1487
        temp[4*i+1]= z1+z2;
1488
        temp[4*i+2]= z1-z2;
1489
        temp[4*i+3]= z0-z3;
1490
    }
1491

    
1492
    for(i=0; i<4; i++){
1493
        const int offset= x_offset[i];
1494
        const int z0= temp[4*0+i] + temp[4*2+i];
1495
        const int z1= temp[4*0+i] - temp[4*2+i];
1496
        const int z2= temp[4*1+i] - temp[4*3+i];
1497
        const int z3= temp[4*1+i] + temp[4*3+i];
1498

    
1499
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
1500
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1501
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1502
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1503
    }
1504
}
1505

    
1506
#if 0
1507
/**
1508
 * DCT transforms the 16 dc values.
1509
 * @param qp quantization parameter ??? FIXME
1510
 */
1511
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1512
//    const int qmul= dequant_coeff[qp][0];
1513
    int i;
1514
    int temp[16]; //FIXME check if this is a good idea
1515
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1516
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1517

1518
    for(i=0; i<4; i++){
1519
        const int offset= y_offset[i];
1520
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1521
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1522
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1523
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1524

1525
        temp[4*i+0]= z0+z3;
1526
        temp[4*i+1]= z1+z2;
1527
        temp[4*i+2]= z1-z2;
1528
        temp[4*i+3]= z0-z3;
1529
    }
1530

1531
    for(i=0; i<4; i++){
1532
        const int offset= x_offset[i];
1533
        const int z0= temp[4*0+i] + temp[4*2+i];
1534
        const int z1= temp[4*0+i] - temp[4*2+i];
1535
        const int z2= temp[4*1+i] - temp[4*3+i];
1536
        const int z3= temp[4*1+i] + temp[4*3+i];
1537

1538
        block[stride*0 +offset]= (z0 + z3)>>1;
1539
        block[stride*2 +offset]= (z1 + z2)>>1;
1540
        block[stride*8 +offset]= (z1 - z2)>>1;
1541
        block[stride*10+offset]= (z0 - z3)>>1;
1542
    }
1543
}
1544
#endif
1545

    
1546
#undef xStride
1547
#undef stride
1548

    
1549
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1550
    const int stride= 16*2;
1551
    const int xStride= 16;
1552
    int a,b,c,d,e;
1553

    
1554
    a= block[stride*0 + xStride*0];
1555
    b= block[stride*0 + xStride*1];
1556
    c= block[stride*1 + xStride*0];
1557
    d= block[stride*1 + xStride*1];
1558

    
1559
    e= a-b;
1560
    a= a+b;
1561
    b= c-d;
1562
    c= c+d;
1563

    
1564
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1565
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1566
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1567
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1568
}
1569

    
1570
#if 0
1571
static void chroma_dc_dct_c(DCTELEM *block){
1572
    const int stride= 16*2;
1573
    const int xStride= 16;
1574
    int a,b,c,d,e;
1575

1576
    a= block[stride*0 + xStride*0];
1577
    b= block[stride*0 + xStride*1];
1578
    c= block[stride*1 + xStride*0];
1579
    d= block[stride*1 + xStride*1];
1580

1581
    e= a-b;
1582
    a= a+b;
1583
    b= c-d;
1584
    c= c+d;
1585

1586
    block[stride*0 + xStride*0]= (a+c);
1587
    block[stride*0 + xStride*1]= (e+b);
1588
    block[stride*1 + xStride*0]= (a-c);
1589
    block[stride*1 + xStride*1]= (e-b);
1590
}
1591
#endif
1592

    
1593
/**
1594
 * gets the chroma qp.
1595
 */
1596
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1597
    return h->pps.chroma_qp_table[t][qscale];
1598
}
1599

    
1600
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1601
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1602
                           int src_x_offset, int src_y_offset,
1603
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1604
    MpegEncContext * const s = &h->s;
1605
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1606
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1607
    const int luma_xy= (mx&3) + ((my&3)<<2);
1608
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1609
    uint8_t * src_cb, * src_cr;
1610
    int extra_width= h->emu_edge_width;
1611
    int extra_height= h->emu_edge_height;
1612
    int emu=0;
1613
    const int full_mx= mx>>2;
1614
    const int full_my= my>>2;
1615
    const int pic_width  = 16*s->mb_width;
1616
    const int pic_height = 16*s->mb_height >> MB_FIELD;
1617

    
1618
    if(mx&7) extra_width -= 3;
1619
    if(my&7) extra_height -= 3;
1620

    
1621
    if(   full_mx < 0-extra_width
1622
       || full_my < 0-extra_height
1623
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
1624
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
1625
        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);
1626
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1627
        emu=1;
1628
    }
1629

    
1630
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1631
    if(!square){
1632
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1633
    }
1634

    
1635
    if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1636

    
1637
    if(MB_FIELD){
1638
        // chroma offset when predicting from a field of opposite parity
1639
        my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1640
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1641
    }
1642
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1643
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1644

    
1645
    if(emu){
1646
        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);
1647
            src_cb= s->edge_emu_buffer;
1648
    }
1649
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1650

    
1651
    if(emu){
1652
        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);
1653
            src_cr= s->edge_emu_buffer;
1654
    }
1655
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1656
}
1657

    
1658
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1659
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1660
                           int x_offset, int y_offset,
1661
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1662
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1663
                           int list0, int list1){
1664
    MpegEncContext * const s = &h->s;
1665
    qpel_mc_func *qpix_op=  qpix_put;
1666
    h264_chroma_mc_func chroma_op= chroma_put;
1667

    
1668
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1669
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1670
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1671
    x_offset += 8*s->mb_x;
1672
    y_offset += 8*(s->mb_y >> MB_FIELD);
1673

    
1674
    if(list0){
1675
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1676
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1677
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1678
                           qpix_op, chroma_op);
1679

    
1680
        qpix_op=  qpix_avg;
1681
        chroma_op= chroma_avg;
1682
    }
1683

    
1684
    if(list1){
1685
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1686
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1687
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1688
                           qpix_op, chroma_op);
1689
    }
1690
}
1691

    
1692
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1693
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1694
                           int x_offset, int y_offset,
1695
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1696
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1697
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1698
                           int list0, int list1){
1699
    MpegEncContext * const s = &h->s;
1700

    
1701
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1702
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1703
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1704
    x_offset += 8*s->mb_x;
1705
    y_offset += 8*(s->mb_y >> MB_FIELD);
1706

    
1707
    if(list0 && list1){
1708
        /* don't optimize for luma-only case, since B-frames usually
1709
         * use implicit weights => chroma too. */
1710
        uint8_t *tmp_cb = s->obmc_scratchpad;
1711
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1712
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1713
        int refn0 = h->ref_cache[0][ scan8[n] ];
1714
        int refn1 = h->ref_cache[1][ scan8[n] ];
1715

    
1716
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1717
                    dest_y, dest_cb, dest_cr,
1718
                    x_offset, y_offset, qpix_put, chroma_put);
1719
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1720
                    tmp_y, tmp_cb, tmp_cr,
1721
                    x_offset, y_offset, qpix_put, chroma_put);
1722

    
1723
        if(h->use_weight == 2){
1724
            int weight0 = h->implicit_weight[refn0][refn1];
1725
            int weight1 = 64 - weight0;
1726
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
1727
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1728
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1729
        }else{
1730
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1731
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1732
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1733
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1734
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1735
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1736
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1737
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1738
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1739
        }
1740
    }else{
1741
        int list = list1 ? 1 : 0;
1742
        int refn = h->ref_cache[list][ scan8[n] ];
1743
        Picture *ref= &h->ref_list[list][refn];
1744
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1745
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
1746
                    qpix_put, chroma_put);
1747

    
1748
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1749
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
1750
        if(h->use_weight_chroma){
1751
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1752
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1753
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1754
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1755
        }
1756
    }
1757
}
1758

    
1759
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1760
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1761
                           int x_offset, int y_offset,
1762
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1763
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1764
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1765
                           int list0, int list1){
1766
    if((h->use_weight==2 && list0 && list1
1767
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1768
       || h->use_weight==1)
1769
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1770
                         x_offset, y_offset, qpix_put, chroma_put,
1771
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1772
    else
1773
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1774
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1775
}
1776

    
1777
static inline void prefetch_motion(H264Context *h, int list){
1778
    /* fetch pixels for estimated mv 4 macroblocks ahead
1779
     * optimized for 64byte cache lines */
1780
    MpegEncContext * const s = &h->s;
1781
    const int refn = h->ref_cache[list][scan8[0]];
1782
    if(refn >= 0){
1783
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1784
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1785
        uint8_t **src= h->ref_list[list][refn].data;
1786
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1787
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
1788
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1789
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1790
    }
1791
}
1792

    
1793
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1794
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1795
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1796
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1797
    MpegEncContext * const s = &h->s;
1798
    const int mb_xy= h->mb_xy;
1799
    const int mb_type= s->current_picture.mb_type[mb_xy];
1800

    
1801
    assert(IS_INTER(mb_type));
1802

    
1803
    prefetch_motion(h, 0);
1804

    
1805
    if(IS_16X16(mb_type)){
1806
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1807
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1808
                &weight_op[0], &weight_avg[0],
1809
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1810
    }else if(IS_16X8(mb_type)){
1811
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1812
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1813
                &weight_op[1], &weight_avg[1],
1814
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1815
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1816
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1817
                &weight_op[1], &weight_avg[1],
1818
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1819
    }else if(IS_8X16(mb_type)){
1820
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1821
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1822
                &weight_op[2], &weight_avg[2],
1823
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1824
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1825
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1826
                &weight_op[2], &weight_avg[2],
1827
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1828
    }else{
1829
        int i;
1830

    
1831
        assert(IS_8X8(mb_type));
1832

    
1833
        for(i=0; i<4; i++){
1834
            const int sub_mb_type= h->sub_mb_type[i];
1835
            const int n= 4*i;
1836
            int x_offset= (i&1)<<2;
1837
            int y_offset= (i&2)<<1;
1838

    
1839
            if(IS_SUB_8X8(sub_mb_type)){
1840
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1841
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1842
                    &weight_op[3], &weight_avg[3],
1843
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1844
            }else if(IS_SUB_8X4(sub_mb_type)){
1845
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1846
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1847
                    &weight_op[4], &weight_avg[4],
1848
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1849
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1850
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1851
                    &weight_op[4], &weight_avg[4],
1852
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1853
            }else if(IS_SUB_4X8(sub_mb_type)){
1854
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1855
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1856
                    &weight_op[5], &weight_avg[5],
1857
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1858
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1859
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1860
                    &weight_op[5], &weight_avg[5],
1861
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1862
            }else{
1863
                int j;
1864
                assert(IS_SUB_4X4(sub_mb_type));
1865
                for(j=0; j<4; j++){
1866
                    int sub_x_offset= x_offset + 2*(j&1);
1867
                    int sub_y_offset= y_offset +   (j&2);
1868
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1869
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1870
                        &weight_op[6], &weight_avg[6],
1871
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1872
                }
1873
            }
1874
        }
1875
    }
1876

    
1877
    prefetch_motion(h, 1);
1878
}
1879

    
1880
static av_cold void init_cavlc_level_tab(void){
1881
    int suffix_length, mask;
1882
    unsigned int i;
1883

    
1884
    for(suffix_length=0; suffix_length<7; suffix_length++){
1885
        for(i=0; i<(1<<LEVEL_TAB_BITS); i++){
1886
            int prefix= LEVEL_TAB_BITS - av_log2(2*i);
1887
            int level_code= (prefix<<suffix_length) + (i>>(LEVEL_TAB_BITS-prefix-1-suffix_length)) - (1<<suffix_length);
1888

    
1889
            mask= -(level_code&1);
1890
            level_code= (((2+level_code)>>1) ^ mask) - mask;
1891
            if(prefix + 1 + suffix_length <= LEVEL_TAB_BITS){
1892
                cavlc_level_tab[suffix_length][i][0]= level_code;
1893
                cavlc_level_tab[suffix_length][i][1]= prefix + 1 + suffix_length;
1894
            }else if(prefix + 1 <= LEVEL_TAB_BITS){
1895
                cavlc_level_tab[suffix_length][i][0]= prefix+100;
1896
                cavlc_level_tab[suffix_length][i][1]= prefix + 1;
1897
            }else{
1898
                cavlc_level_tab[suffix_length][i][0]= LEVEL_TAB_BITS+100;
1899
                cavlc_level_tab[suffix_length][i][1]= LEVEL_TAB_BITS;
1900
            }
1901
        }
1902
    }
1903
}
1904

    
1905
static av_cold void decode_init_vlc(void){
1906
    static int done = 0;
1907

    
1908
    if (!done) {
1909
        int i;
1910
        int offset;
1911
        done = 1;
1912

    
1913
        chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
1914
        chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
1915
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1916
                 &chroma_dc_coeff_token_len [0], 1, 1,
1917
                 &chroma_dc_coeff_token_bits[0], 1, 1,
1918
                 INIT_VLC_USE_NEW_STATIC);
1919

    
1920
        offset = 0;
1921
        for(i=0; i<4; i++){
1922
            coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
1923
            coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
1924
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1925
                     &coeff_token_len [i][0], 1, 1,
1926
                     &coeff_token_bits[i][0], 1, 1,
1927
                     INIT_VLC_USE_NEW_STATIC);
1928
            offset += coeff_token_vlc_tables_size[i];
1929
        }
1930
        /*
1931
         * This is a one time safety check to make sure that
1932
         * the packed static coeff_token_vlc table sizes
1933
         * were initialized correctly.
1934
         */
1935
        assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
1936

    
1937
        for(i=0; i<3; i++){
1938
            chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
1939
            chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
1940
            init_vlc(&chroma_dc_total_zeros_vlc[i],
1941
                     CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1942
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
1943
                     &chroma_dc_total_zeros_bits[i][0], 1, 1,
1944
                     INIT_VLC_USE_NEW_STATIC);
1945
        }
1946
        for(i=0; i<15; i++){
1947
            total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
1948
            total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
1949
            init_vlc(&total_zeros_vlc[i],
1950
                     TOTAL_ZEROS_VLC_BITS, 16,
1951
                     &total_zeros_len [i][0], 1, 1,
1952
                     &total_zeros_bits[i][0], 1, 1,
1953
                     INIT_VLC_USE_NEW_STATIC);
1954
        }
1955

    
1956
        for(i=0; i<6; i++){
1957
            run_vlc[i].table = run_vlc_tables[i];
1958
            run_vlc[i].table_allocated = run_vlc_tables_size;
1959
            init_vlc(&run_vlc[i],
1960
                     RUN_VLC_BITS, 7,
1961
                     &run_len [i][0], 1, 1,
1962
                     &run_bits[i][0], 1, 1,
1963
                     INIT_VLC_USE_NEW_STATIC);
1964
        }
1965
        run7_vlc.table = run7_vlc_table,
1966
        run7_vlc.table_allocated = run7_vlc_table_size;
1967
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1968
                 &run_len [6][0], 1, 1,
1969
                 &run_bits[6][0], 1, 1,
1970
                 INIT_VLC_USE_NEW_STATIC);
1971

    
1972
        init_cavlc_level_tab();
1973
    }
1974
}
1975

    
1976
static void free_tables(H264Context *h){
1977
    int i;
1978
    H264Context *hx;
1979
    av_freep(&h->intra4x4_pred_mode);
1980
    av_freep(&h->chroma_pred_mode_table);
1981
    av_freep(&h->cbp_table);
1982
    av_freep(&h->mvd_table[0]);
1983
    av_freep(&h->mvd_table[1]);
1984
    av_freep(&h->direct_table);
1985
    av_freep(&h->non_zero_count);
1986
    av_freep(&h->slice_table_base);
1987
    h->slice_table= NULL;
1988

    
1989
    av_freep(&h->mb2b_xy);
1990
    av_freep(&h->mb2b8_xy);
1991

    
1992
    for(i = 0; i < h->s.avctx->thread_count; i++) {
1993
        hx = h->thread_context[i];
1994
        if(!hx) continue;
1995
        av_freep(&hx->top_borders[1]);
1996
        av_freep(&hx->top_borders[0]);
1997
        av_freep(&hx->s.obmc_scratchpad);
1998
    }
1999
}
2000

    
2001
static void init_dequant8_coeff_table(H264Context *h){
2002
    int i,q,x;
2003
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2004
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
2005
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
2006

    
2007
    for(i=0; i<2; i++ ){
2008
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2009
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
2010
            break;
2011
        }
2012

    
2013
        for(q=0; q<52; q++){
2014
            int shift = div6[q];
2015
            int idx = rem6[q];
2016
            for(x=0; x<64; x++)
2017
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2018
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2019
                    h->pps.scaling_matrix8[i][x]) << shift;
2020
        }
2021
    }
2022
}
2023

    
2024
static void init_dequant4_coeff_table(H264Context *h){
2025
    int i,j,q,x;
2026
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2027
    for(i=0; i<6; i++ ){
2028
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
2029
        for(j=0; j<i; j++){
2030
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2031
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2032
                break;
2033
            }
2034
        }
2035
        if(j<i)
2036
            continue;
2037

    
2038
        for(q=0; q<52; q++){
2039
            int shift = div6[q] + 2;
2040
            int idx = rem6[q];
2041
            for(x=0; x<16; x++)
2042
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2043
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2044
                    h->pps.scaling_matrix4[i][x]) << shift;
2045
        }
2046
    }
2047
}
2048

    
2049
static void init_dequant_tables(H264Context *h){
2050
    int i,x;
2051
    init_dequant4_coeff_table(h);
2052
    if(h->pps.transform_8x8_mode)
2053
        init_dequant8_coeff_table(h);
2054
    if(h->sps.transform_bypass){
2055
        for(i=0; i<6; i++)
2056
            for(x=0; x<16; x++)
2057
                h->dequant4_coeff[i][0][x] = 1<<6;
2058
        if(h->pps.transform_8x8_mode)
2059
            for(i=0; i<2; i++)
2060
                for(x=0; x<64; x++)
2061
                    h->dequant8_coeff[i][0][x] = 1<<6;
2062
    }
2063
}
2064

    
2065

    
2066
/**
2067
 * allocates tables.
2068
 * needs width/height
2069
 */
2070
static int alloc_tables(H264Context *h){
2071
    MpegEncContext * const s = &h->s;
2072
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2073
    int x,y;
2074

    
2075
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2076

    
2077
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2078
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base))
2079
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2080

    
2081
    CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2082
    CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2083
    CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2084
    CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2085

    
2086
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
2087
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2088

    
2089
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2090
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2091
    for(y=0; y<s->mb_height; y++){
2092
        for(x=0; x<s->mb_width; x++){
2093
            const int mb_xy= x + y*s->mb_stride;
2094
            const int b_xy = 4*x + 4*y*h->b_stride;
2095
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2096

    
2097
            h->mb2b_xy [mb_xy]= b_xy;
2098
            h->mb2b8_xy[mb_xy]= b8_xy;
2099
        }
2100
    }
2101

    
2102
    s->obmc_scratchpad = NULL;
2103

    
2104
    if(!h->dequant4_coeff[0])
2105
        init_dequant_tables(h);
2106

    
2107
    return 0;
2108
fail:
2109
    free_tables(h);
2110
    return -1;
2111
}
2112

    
2113
/**
2114
 * Mimic alloc_tables(), but for every context thread.
2115
 */
2116
static void clone_tables(H264Context *dst, H264Context *src){
2117
    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
2118
    dst->non_zero_count           = src->non_zero_count;
2119
    dst->slice_table              = src->slice_table;
2120
    dst->cbp_table                = src->cbp_table;
2121
    dst->mb2b_xy                  = src->mb2b_xy;
2122
    dst->mb2b8_xy                 = src->mb2b8_xy;
2123
    dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
2124
    dst->mvd_table[0]             = src->mvd_table[0];
2125
    dst->mvd_table[1]             = src->mvd_table[1];
2126
    dst->direct_table             = src->direct_table;
2127

    
2128
    dst->s.obmc_scratchpad = NULL;
2129
    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2130
}
2131

    
2132
/**
2133
 * Init context
2134
 * Allocate buffers which are not shared amongst multiple threads.
2135
 */
2136
static int context_init(H264Context *h){
2137
    CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2138
    CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2139

    
2140
    return 0;
2141
fail:
2142
    return -1; // free_tables will clean up for us
2143
}
2144

    
2145
static av_cold void common_init(H264Context *h){
2146
    MpegEncContext * const s = &h->s;
2147

    
2148
    s->width = s->avctx->width;
2149
    s->height = s->avctx->height;
2150
    s->codec_id= s->avctx->codec->id;
2151

    
2152
    ff_h264_pred_init(&h->hpc, s->codec_id);
2153

    
2154
    h->dequant_coeff_pps= -1;
2155
    s->unrestricted_mv=1;
2156
    s->decode=1; //FIXME
2157

    
2158
    dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
2159

    
2160
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2161
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2162
}
2163

    
2164
static av_cold int decode_init(AVCodecContext *avctx){
2165
    H264Context *h= avctx->priv_data;
2166
    MpegEncContext * const s = &h->s;
2167

    
2168
    MPV_decode_defaults(s);
2169

    
2170
    s->avctx = avctx;
2171
    common_init(h);
2172

    
2173
    s->out_format = FMT_H264;
2174
    s->workaround_bugs= avctx->workaround_bugs;
2175

    
2176
    // set defaults
2177
//    s->decode_mb= ff_h263_decode_mb;
2178
    s->quarter_sample = 1;
2179
    s->low_delay= 1;
2180

    
2181
    if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2182
        avctx->pix_fmt= PIX_FMT_VDPAU_H264;
2183
    else
2184
        avctx->pix_fmt= avctx->get_format(avctx, avctx->codec->pix_fmts);
2185
    avctx->hwaccel = ff_find_hwaccel(avctx->codec->id, avctx->pix_fmt);
2186

    
2187
    decode_init_vlc();
2188

    
2189
    if(avctx->extradata_size > 0 && avctx->extradata &&
2190
       *(char *)avctx->extradata == 1){
2191
        h->is_avc = 1;
2192
        h->got_avcC = 0;
2193
    } else {
2194
        h->is_avc = 0;
2195
    }
2196

    
2197
    h->thread_context[0] = h;
2198
    h->outputed_poc = INT_MIN;
2199
    h->prev_poc_msb= 1<<16;
2200
    h->sei_recovery_frame_cnt = -1;
2201
    h->sei_dpb_output_delay = 0;
2202
    h->sei_cpb_removal_delay = -1;
2203
    h->sei_buffering_period_present = 0;
2204
    if(avctx->codec_id == CODEC_ID_H264)
2205
        avctx->ticks_per_frame = 2;
2206
    return 0;
2207
}
2208

    
2209
static int frame_start(H264Context *h){
2210
    MpegEncContext * const s = &h->s;
2211
    int i;
2212

    
2213
    if(MPV_frame_start(s, s->avctx) < 0)
2214
        return -1;
2215
    ff_er_frame_start(s);
2216
    /*
2217
     * MPV_frame_start uses pict_type to derive key_frame.
2218
     * This is incorrect for H.264; IDR markings must be used.
2219
     * Zero here; IDR markings per slice in frame or fields are ORed in later.
2220
     * See decode_nal_units().
2221
     */
2222
    s->current_picture_ptr->key_frame= 0;
2223

    
2224
    assert(s->linesize && s->uvlinesize);
2225

    
2226
    for(i=0; i<16; i++){
2227
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2228
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2229
    }
2230
    for(i=0; i<4; i++){
2231
        h->block_offset[16+i]=
2232
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2233
        h->block_offset[24+16+i]=
2234
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2235
    }
2236

    
2237
    /* can't be in alloc_tables because linesize isn't known there.
2238
     * FIXME: redo bipred weight to not require extra buffer? */
2239
    for(i = 0; i < s->avctx->thread_count; i++)
2240
        if(!h->thread_context[i]->s.obmc_scratchpad)
2241
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2242

    
2243
    /* some macroblocks will be accessed before they're available */
2244
    if(FRAME_MBAFF || s->avctx->thread_count > 1)
2245
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
2246

    
2247
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2248

    
2249
    // We mark the current picture as non-reference after allocating it, so
2250
    // that if we break out due to an error it can be released automatically
2251
    // in the next MPV_frame_start().
2252
    // SVQ3 as well as most other codecs have only last/next/current and thus
2253
    // get released even with set reference, besides SVQ3 and others do not
2254
    // mark frames as reference later "naturally".
2255
    if(s->codec_id != CODEC_ID_SVQ3)
2256
        s->current_picture_ptr->reference= 0;
2257

    
2258
    s->current_picture_ptr->field_poc[0]=
2259
    s->current_picture_ptr->field_poc[1]= INT_MAX;
2260
    assert(s->current_picture_ptr->long_ref==0);
2261

    
2262
    return 0;
2263
}
2264

    
2265
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){
2266
    MpegEncContext * const s = &h->s;
2267
    int i;
2268
    int step    = 1;
2269
    int offset  = 1;
2270
    int uvoffset= 1;
2271
    int top_idx = 1;
2272
    int skiplast= 0;
2273

    
2274
    src_y  -=   linesize;
2275
    src_cb -= uvlinesize;
2276
    src_cr -= uvlinesize;
2277

    
2278
    if(!simple && FRAME_MBAFF){
2279
        if(s->mb_y&1){
2280
            offset  = MB_MBAFF ? 1 : 17;
2281
            uvoffset= MB_MBAFF ? 1 : 9;
2282
            if(!MB_MBAFF){
2283
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y +  15*linesize);
2284
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
2285
                if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2286
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
2287
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
2288
                }
2289
            }
2290
        }else{
2291
            if(!MB_MBAFF){
2292
                h->left_border[0]= h->top_borders[0][s->mb_x][15];
2293
                if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2294
                    h->left_border[34   ]= h->top_borders[0][s->mb_x][16+7  ];
2295
                    h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
2296
                }
2297
                skiplast= 1;
2298
            }
2299
            offset  =
2300
            uvoffset=
2301
            top_idx = MB_MBAFF ? 0 : 1;
2302
        }
2303
        step= MB_MBAFF ? 2 : 1;
2304
    }
2305

    
2306
    // There are two lines saved, the line above the the top macroblock of a pair,
2307
    // and the line above the bottom macroblock
2308
    h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
2309
    for(i=1; i<17 - skiplast; i++){
2310
        h->left_border[offset+i*step]= src_y[15+i*  linesize];
2311
    }
2312

    
2313
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2314
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2315

    
2316
    if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2317
        h->left_border[uvoffset+34   ]= h->top_borders[top_idx][s->mb_x][16+7];
2318
        h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
2319
        for(i=1; i<9 - skiplast; i++){
2320
            h->left_border[uvoffset+34   +i*step]= src_cb[7+i*uvlinesize];
2321
            h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
2322
        }
2323
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2324
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2325
    }
2326
}
2327

    
2328
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){
2329
    MpegEncContext * const s = &h->s;
2330
    int temp8, i;
2331
    uint64_t temp64;
2332
    int deblock_left;
2333
    int deblock_top;
2334
    int mb_xy;
2335
    int step    = 1;
2336
    int offset  = 1;
2337
    int uvoffset= 1;
2338
    int top_idx = 1;
2339

    
2340
    if(!simple && FRAME_MBAFF){
2341
        if(s->mb_y&1){
2342
            offset  = MB_MBAFF ? 1 : 17;
2343
            uvoffset= MB_MBAFF ? 1 : 9;
2344
        }else{
2345
            offset  =
2346
            uvoffset=
2347
            top_idx = MB_MBAFF ? 0 : 1;
2348
        }
2349
        step= MB_MBAFF ? 2 : 1;
2350
    }
2351

    
2352
    if(h->deblocking_filter == 2) {
2353
        mb_xy = h->mb_xy;
2354
        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2355
        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2356
    } else {
2357
        deblock_left = (s->mb_x > 0);
2358
        deblock_top =  (s->mb_y > !!MB_FIELD);
2359
    }
2360

    
2361
    src_y  -=   linesize + 1;
2362
    src_cb -= uvlinesize + 1;
2363
    src_cr -= uvlinesize + 1;
2364

    
2365
#define XCHG(a,b,t,xchg)\
2366
t= a;\
2367
if(xchg)\
2368
    a= b;\
2369
b= t;
2370

    
2371
    if(deblock_left){
2372
        for(i = !deblock_top; i<16; i++){
2373
            XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, xchg);
2374
        }
2375
        XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, 1);
2376
    }
2377

    
2378
    if(deblock_top){
2379
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2380
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2381
        if(s->mb_x+1 < s->mb_width){
2382
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2383
        }
2384
    }
2385

    
2386
    if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2387
        if(deblock_left){
2388
            for(i = !deblock_top; i<8; i++){
2389
                XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, xchg);
2390
                XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
2391
            }
2392
            XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, 1);
2393
            XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
2394
        }
2395
        if(deblock_top){
2396
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2397
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2398
        }
2399
    }
2400
}
2401

    
2402
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2403
    MpegEncContext * const s = &h->s;
2404
    const int mb_x= s->mb_x;
2405
    const int mb_y= s->mb_y;
2406
    const int mb_xy= h->mb_xy;
2407
    const int mb_type= s->current_picture.mb_type[mb_xy];
2408
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2409
    int linesize, uvlinesize /*dct_offset*/;
2410
    int i;
2411
    int *block_offset = &h->block_offset[0];
2412
    const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2413
    /* is_h264 should always be true if SVQ3 is disabled. */
2414
    const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
2415
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2416
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2417

    
2418
    dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2419
    dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2420
    dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2421

    
2422
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2423
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2424

    
2425
    if (!simple && MB_FIELD) {
2426
        linesize   = h->mb_linesize   = s->linesize * 2;
2427
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2428
        block_offset = &h->block_offset[24];
2429
        if(mb_y&1){ //FIXME move out of this function?
2430
            dest_y -= s->linesize*15;
2431
            dest_cb-= s->uvlinesize*7;
2432
            dest_cr-= s->uvlinesize*7;
2433
        }
2434
        if(FRAME_MBAFF) {
2435
            int list;
2436
            for(list=0; list<h->list_count; list++){
2437
                if(!USES_LIST(mb_type, list))
2438
                    continue;
2439
                if(IS_16X16(mb_type)){
2440
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
2441
                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2442
                }else{
2443
                    for(i=0; i<16; i+=4){
2444
                        int ref = h->ref_cache[list][scan8[i]];
2445
                        if(ref >= 0)
2446
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2447
                    }
2448
                }
2449
            }
2450
        }
2451
    } else {
2452
        linesize   = h->mb_linesize   = s->linesize;
2453
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2454
//        dct_offset = s->linesize * 16;
2455
    }
2456

    
2457
    if (!simple && IS_INTRA_PCM(mb_type)) {
2458
        for (i=0; i<16; i++) {
2459
            memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
2460
        }
2461
        for (i=0; i<8; i++) {
2462
            memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
2463
            memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
2464
        }
2465
    } else {
2466
        if(IS_INTRA(mb_type)){
2467
            if(h->deblocking_filter)
2468
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2469

    
2470
            if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2471
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2472
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2473
            }
2474

    
2475
            if(IS_INTRA4x4(mb_type)){
2476
                if(simple || !s->encoding){
2477
                    if(IS_8x8DCT(mb_type)){
2478
                        if(transform_bypass){
2479
                            idct_dc_add =
2480
                            idct_add    = s->dsp.add_pixels8;
2481
                        }else{
2482
                            idct_dc_add = s->dsp.h264_idct8_dc_add;
2483
                            idct_add    = s->dsp.h264_idct8_add;
2484
                        }
2485
                        for(i=0; i<16; i+=4){
2486
                            uint8_t * const ptr= dest_y + block_offset[i];
2487
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2488
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2489
                                h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
2490
                            }else{
2491
                                const int nnz = h->non_zero_count_cache[ scan8[i] ];
2492
                                h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2493
                                                            (h->topright_samples_available<<i)&0x4000, linesize);
2494
                                if(nnz){
2495
                                    if(nnz == 1 && h->mb[i*16])
2496
                                        idct_dc_add(ptr, h->mb + i*16, linesize);
2497
                                    else
2498
                                        idct_add   (ptr, h->mb + i*16, linesize);
2499
                                }
2500
                            }
2501
                        }
2502
                    }else{
2503
                        if(transform_bypass){
2504
                            idct_dc_add =
2505
                            idct_add    = s->dsp.add_pixels4;
2506
                        }else{
2507
                            idct_dc_add = s->dsp.h264_idct_dc_add;
2508
                            idct_add    = s->dsp.h264_idct_add;
2509
                        }
2510
                        for(i=0; i<16; i++){
2511
                            uint8_t * const ptr= dest_y + block_offset[i];
2512
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2513

    
2514
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2515
                                h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
2516
                            }else{
2517
                                uint8_t *topright;
2518
                                int nnz, tr;
2519
                                if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2520
                                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2521
                                    assert(mb_y || linesize <= block_offset[i]);
2522
                                    if(!topright_avail){
2523
                                        tr= ptr[3 - linesize]*0x01010101;
2524
                                        topright= (uint8_t*) &tr;
2525
                                    }else
2526
                                        topright= ptr + 4 - linesize;
2527
                                }else
2528
                                    topright= NULL;
2529

    
2530
                                h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2531
                                nnz = h->non_zero_count_cache[ scan8[i] ];
2532
                                if(nnz){
2533
                                    if(is_h264){
2534
                                        if(nnz == 1 && h->mb[i*16])
2535
                                            idct_dc_add(ptr, h->mb + i*16, linesize);
2536
                                        else
2537
                                            idct_add   (ptr, h->mb + i*16, linesize);
2538
                                    }else
2539
                                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2540
                                }
2541
                            }
2542
                        }
2543
                    }
2544
                }
2545
            }else{
2546
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2547
                if(is_h264){
2548
                    if(!transform_bypass)
2549
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
2550
                }else
2551
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2552
            }
2553
            if(h->deblocking_filter)
2554
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2555
        }else if(is_h264){
2556
            hl_motion(h, dest_y, dest_cb, dest_cr,
2557
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2558
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2559
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2560
        }
2561

    
2562

    
2563
        if(!IS_INTRA4x4(mb_type)){
2564
            if(is_h264){
2565
                if(IS_INTRA16x16(mb_type)){
2566
                    if(transform_bypass){
2567
                        if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
2568
                            h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
2569
                        }else{
2570
                            for(i=0; i<16; i++){
2571
                                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2572
                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
2573
                            }
2574
                        }
2575
                    }else{
2576
                         s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2577
                    }
2578
                }else if(h->cbp&15){
2579
                    if(transform_bypass){
2580
                        const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2581
                        idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2582
                        for(i=0; i<16; i+=di){
2583
                            if(h->non_zero_count_cache[ scan8[i] ]){
2584
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2585
                            }
2586
                        }
2587
                    }else{
2588
                        if(IS_8x8DCT(mb_type)){
2589
                            s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2590
                        }else{
2591
                            s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2592
                        }
2593
                    }
2594
                }
2595
            }else{
2596
                for(i=0; i<16; i++){
2597
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2598
                        uint8_t * const ptr= dest_y + block_offset[i];
2599
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2600
                    }
2601
                }
2602
            }
2603
        }
2604

    
2605
        if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2606
            uint8_t *dest[2] = {dest_cb, dest_cr};
2607
            if(transform_bypass){
2608
                if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2609
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
2610
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
2611
                }else{
2612
                    idct_add = s->dsp.add_pixels4;
2613
                    for(i=16; i<16+8; i++){
2614
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2615
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2616
                    }
2617
                }
2618
            }else{
2619
                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]);
2620
                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]);
2621
                if(is_h264){
2622
                    idct_add = s->dsp.h264_idct_add;
2623
                    idct_dc_add = s->dsp.h264_idct_dc_add;
2624
                    for(i=16; i<16+8; i++){
2625
                        if(h->non_zero_count_cache[ scan8[i] ])
2626
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2627
                        else if(h->mb[i*16])
2628
                            idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2629
                    }
2630
                }else{
2631
                    for(i=16; i<16+8; i++){
2632
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2633
                            uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
2634
                            svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2635
                        }
2636
                    }
2637
                }
2638
            }
2639
        }
2640
    }
2641
    if(h->cbp || IS_INTRA(mb_type))
2642
        s->dsp.clear_blocks(h->mb);
2643

    
2644
    if(h->deblocking_filter) {
2645
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
2646
        fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
2647
        h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2648
        h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2649
        if (!simple && FRAME_MBAFF) {
2650
            filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2651
        } else {
2652
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2653
        }
2654
    }
2655
}
2656

    
2657
/**
2658
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2659
 */
2660
static void hl_decode_mb_simple(H264Context *h){
2661
    hl_decode_mb_internal(h, 1);
2662
}
2663

    
2664
/**
2665
 * Process a macroblock; this handles edge cases, such as interlacing.
2666
 */
2667
static void av_noinline hl_decode_mb_complex(H264Context *h){
2668
    hl_decode_mb_internal(h, 0);
2669
}
2670

    
2671
static void hl_decode_mb(H264Context *h){
2672
    MpegEncContext * const s = &h->s;
2673
    const int mb_xy= h->mb_xy;
2674
    const int mb_type= s->current_picture.mb_type[mb_xy];
2675
    int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2676

    
2677
    if (is_complex)
2678
        hl_decode_mb_complex(h);
2679
    else hl_decode_mb_simple(h);
2680
}
2681

    
2682
static void pic_as_field(Picture *pic, const int parity){
2683
    int i;
2684
    for (i = 0; i < 4; ++i) {
2685
        if (parity == PICT_BOTTOM_FIELD)
2686
            pic->data[i] += pic->linesize[i];
2687
        pic->reference = parity;
2688
        pic->linesize[i] *= 2;
2689
    }
2690
    pic->poc= pic->field_poc[parity == PICT_BOTTOM_FIELD];
2691
}
2692

    
2693
static int split_field_copy(Picture *dest, Picture *src,
2694
                            int parity, int id_add){
2695
    int match = !!(src->reference & parity);
2696

    
2697
    if (match) {
2698
        *dest = *src;
2699
        if(parity != PICT_FRAME){
2700
            pic_as_field(dest, parity);
2701
            dest->pic_id *= 2;
2702
            dest->pic_id += id_add;
2703
        }
2704
    }
2705

    
2706
    return match;
2707
}
2708

    
2709
static int build_def_list(Picture *def, Picture **in, int len, int is_long, int sel){
2710
    int i[2]={0};
2711
    int index=0;
2712

    
2713
    while(i[0]<len || i[1]<len){
2714
        while(i[0]<len && !(in[ i[0] ] && (in[ i[0] ]->reference & sel)))
2715
            i[0]++;
2716
        while(i[1]<len && !(in[ i[1] ] && (in[ i[1] ]->reference & (sel^3))))
2717
            i[1]++;
2718
        if(i[0] < len){
2719
            in[ i[0] ]->pic_id= is_long ? i[0] : in[ i[0] ]->frame_num;
2720
            split_field_copy(&def[index++], in[ i[0]++ ], sel  , 1);
2721
        }
2722
        if(i[1] < len){
2723
            in[ i[1] ]->pic_id= is_long ? i[1] : in[ i[1] ]->frame_num;
2724
            split_field_copy(&def[index++], in[ i[1]++ ], sel^3, 0);
2725
        }
2726
    }
2727

    
2728
    return index;
2729
}
2730

    
2731
static int add_sorted(Picture **sorted, Picture **src, int len, int limit, int dir){
2732
    int i, best_poc;
2733
    int out_i= 0;
2734

    
2735
    for(;;){
2736
        best_poc= dir ? INT_MIN : INT_MAX;
2737

    
2738
        for(i=0; i<len; i++){
2739
            const int poc= src[i]->poc;
2740
            if(((poc > limit) ^ dir) && ((poc < best_poc) ^ dir)){
2741
                best_poc= poc;
2742
                sorted[out_i]= src[i];
2743
            }
2744
        }
2745
        if(best_poc == (dir ? INT_MIN : INT_MAX))
2746
            break;
2747
        limit= sorted[out_i++]->poc - dir;
2748
    }
2749
    return out_i;
2750
}
2751

    
2752
/**
2753
 * fills the default_ref_list.
2754
 */
2755
static int fill_default_ref_list(H264Context *h){
2756
    MpegEncContext * const s = &h->s;
2757
    int i, len;
2758

    
2759
    if(h->slice_type_nos==FF_B_TYPE){
2760
        Picture *sorted[32];
2761
        int cur_poc, list;
2762
        int lens[2];
2763

    
2764
        if(FIELD_PICTURE)
2765
            cur_poc= s->current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
2766
        else
2767
            cur_poc= s->current_picture_ptr->poc;
2768

    
2769
        for(list= 0; list<2; list++){
2770
            len= add_sorted(sorted    , h->short_ref, h->short_ref_count, cur_poc, 1^list);
2771
            len+=add_sorted(sorted+len, h->short_ref, h->short_ref_count, cur_poc, 0^list);
2772
            assert(len<=32);
2773
            len= build_def_list(h->default_ref_list[list]    , sorted     , len, 0, s->picture_structure);
2774
            len+=build_def_list(h->default_ref_list[list]+len, h->long_ref, 16 , 1, s->picture_structure);
2775
            assert(len<=32);
2776

    
2777
            if(len < h->ref_count[list])
2778
                memset(&h->default_ref_list[list][len], 0, sizeof(Picture)*(h->ref_count[list] - len));
2779
            lens[list]= len;
2780
        }
2781

    
2782
        if(lens[0] == lens[1] && lens[1] > 1){
2783
            for(i=0; h->default_ref_list[0][i].data[0] == h->default_ref_list[1][i].data[0] && i<lens[0]; i++);
2784
            if(i == lens[0])
2785
                FFSWAP(Picture, h->default_ref_list[1][0], h->default_ref_list[1][1]);
2786
        }
2787
    }else{
2788
        len = build_def_list(h->default_ref_list[0]    , h->short_ref, h->short_ref_count, 0, s->picture_structure);
2789
        len+= build_def_list(h->default_ref_list[0]+len, h-> long_ref, 16                , 1, s->picture_structure);
2790
        assert(len <= 32);
2791
        if(len < h->ref_count[0])
2792
            memset(&h->default_ref_list[0][len], 0, sizeof(Picture)*(h->ref_count[0] - len));
2793
    }
2794
#ifdef TRACE
2795
    for (i=0; i<h->ref_count[0]; i++) {
2796
        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]);
2797
    }
2798
    if(h->slice_type_nos==FF_B_TYPE){
2799
        for (i=0; i<h->ref_count[1]; i++) {
2800
            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]);
2801
        }
2802
    }
2803
#endif
2804
    return 0;
2805
}
2806

    
2807
static void print_short_term(H264Context *h);
2808
static void print_long_term(H264Context *h);
2809

    
2810
/**
2811
 * Extract structure information about the picture described by pic_num in
2812
 * the current decoding context (frame or field). Note that pic_num is
2813
 * picture number without wrapping (so, 0<=pic_num<max_pic_num).
2814
 * @param pic_num picture number for which to extract structure information
2815
 * @param structure one of PICT_XXX describing structure of picture
2816
 *                      with pic_num
2817
 * @return frame number (short term) or long term index of picture
2818
 *         described by pic_num
2819
 */
2820
static int pic_num_extract(H264Context *h, int pic_num, int *structure){
2821
    MpegEncContext * const s = &h->s;
2822

    
2823
    *structure = s->picture_structure;
2824
    if(FIELD_PICTURE){
2825
        if (!(pic_num & 1))
2826
            /* opposite field */
2827
            *structure ^= PICT_FRAME;
2828
        pic_num >>= 1;
2829
    }
2830

    
2831
    return pic_num;
2832
}
2833

    
2834
static int decode_ref_pic_list_reordering(H264Context *h){
2835
    MpegEncContext * const s = &h->s;
2836
    int list, index, pic_structure;
2837

    
2838
    print_short_term(h);
2839
    print_long_term(h);
2840

    
2841
    for(list=0; list<h->list_count; list++){
2842
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2843

    
2844
        if(get_bits1(&s->gb)){
2845
            int pred= h->curr_pic_num;
2846

    
2847
            for(index=0; ; index++){
2848
                unsigned int reordering_of_pic_nums_idc= get_ue_golomb_31(&s->gb);
2849
                unsigned int pic_id;
2850
                int i;
2851
                Picture *ref = NULL;
2852

    
2853
                if(reordering_of_pic_nums_idc==3)
2854
                    break;
2855

    
2856
                if(index >= h->ref_count[list]){
2857
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2858
                    return -1;
2859
                }
2860

    
2861
                if(reordering_of_pic_nums_idc<3){
2862
                    if(reordering_of_pic_nums_idc<2){
2863
                        const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2864
                        int frame_num;
2865

    
2866
                        if(abs_diff_pic_num > h->max_pic_num){
2867
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2868
                            return -1;
2869
                        }
2870

    
2871
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2872
                        else                                pred+= abs_diff_pic_num;
2873
                        pred &= h->max_pic_num - 1;
2874

    
2875
                        frame_num = pic_num_extract(h, pred, &pic_structure);
2876

    
2877
                        for(i= h->short_ref_count-1; i>=0; i--){
2878
                            ref = h->short_ref[i];
2879
                            assert(ref->reference);
2880
                            assert(!ref->long_ref);
2881
                            if(
2882
                                   ref->frame_num == frame_num &&
2883
                                   (ref->reference & pic_structure)
2884
                              )
2885
                                break;
2886
                        }
2887
                        if(i>=0)
2888
                            ref->pic_id= pred;
2889
                    }else{
2890
                        int long_idx;
2891
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2892

    
2893
                        long_idx= pic_num_extract(h, pic_id, &pic_structure);
2894

    
2895
                        if(long_idx>31){
2896
                            av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
2897
                            return -1;
2898
                        }
2899
                        ref = h->long_ref[long_idx];
2900
                        assert(!(ref && !ref->reference));
2901
                        if(ref && (ref->reference & pic_structure)){
2902
                            ref->pic_id= pic_id;
2903
                            assert(ref->long_ref);
2904
                            i=0;
2905
                        }else{
2906
                            i=-1;
2907
                        }
2908
                    }
2909

    
2910
                    if (i < 0) {
2911
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2912
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2913
                    } else {
2914
                        for(i=index; i+1<h->ref_count[list]; i++){
2915
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
2916
                                break;
2917
                        }
2918
                        for(; i > index; i--){
2919
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2920
                        }
2921
                        h->ref_list[list][index]= *ref;
2922
                        if (FIELD_PICTURE){
2923
                            pic_as_field(&h->ref_list[list][index], pic_structure);
2924
                        }
2925
                    }
2926
                }else{
2927
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2928
                    return -1;
2929
                }
2930
            }
2931
        }
2932
    }
2933
    for(list=0; list<h->list_count; list++){
2934
        for(index= 0; index < h->ref_count[list]; index++){
2935
            if(!h->ref_list[list][index].data[0]){
2936
                av_log(h->s.avctx, AV_LOG_ERROR, "Missing reference picture\n");
2937
                h->ref_list[list][index]= s->current_picture; //FIXME this is not a sensible solution
2938
            }
2939
        }
2940
    }
2941

    
2942
    return 0;
2943
}
2944

    
2945
static void fill_mbaff_ref_list(H264Context *h){
2946
    int list, i, j;
2947
    for(list=0; list<2; list++){ //FIXME try list_count
2948
        for(i=0; i<h->ref_count[list]; i++){
2949
            Picture *frame = &h->ref_list[list][i];
2950
            Picture *field = &h->ref_list[list][16+2*i];
2951
            field[0] = *frame;
2952
            for(j=0; j<3; j++)
2953
                field[0].linesize[j] <<= 1;
2954
            field[0].reference = PICT_TOP_FIELD;
2955
            field[0].poc= field[0].field_poc[0];
2956
            field[1] = field[0];
2957
            for(j=0; j<3; j++)
2958
                field[1].data[j] += frame->linesize[j];
2959
            field[1].reference = PICT_BOTTOM_FIELD;
2960
            field[1].poc= field[1].field_poc[1];
2961

    
2962
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
2963
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
2964
            for(j=0; j<2; j++){
2965
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
2966
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
2967
            }
2968
        }
2969
    }
2970
    for(j=0; j<h->ref_count[1]; j++){
2971
        for(i=0; i<h->ref_count[0]; i++)
2972
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
2973
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
2974
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
2975
    }
2976
}
2977

    
2978
static int pred_weight_table(H264Context *h){
2979
    MpegEncContext * const s = &h->s;
2980
    int list, i;
2981
    int luma_def, chroma_def;
2982

    
2983
    h->use_weight= 0;
2984
    h->use_weight_chroma= 0;
2985
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2986
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2987
    luma_def = 1<<h->luma_log2_weight_denom;
2988
    chroma_def = 1<<h->chroma_log2_weight_denom;
2989

    
2990
    for(list=0; list<2; list++){
2991
        h->luma_weight_flag[list]   = 0;
2992
        h->chroma_weight_flag[list] = 0;
2993
        for(i=0; i<h->ref_count[list]; i++){
2994
            int luma_weight_flag, chroma_weight_flag;
2995

    
2996
            luma_weight_flag= get_bits1(&s->gb);
2997
            if(luma_weight_flag){
2998
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
2999
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3000
                if(   h->luma_weight[list][i] != luma_def
3001
                   || h->luma_offset[list][i] != 0) {
3002
                    h->use_weight= 1;
3003
                    h->luma_weight_flag[list]= 1;
3004
                }
3005
            }else{
3006
                h->luma_weight[list][i]= luma_def;
3007
                h->luma_offset[list][i]= 0;
3008
            }
3009

    
3010
            if(CHROMA){
3011
                chroma_weight_flag= get_bits1(&s->gb);
3012
                if(chroma_weight_flag){
3013
                    int j;
3014
                    for(j=0; j<2; j++){
3015
                        h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3016
                        h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3017
                        if(   h->chroma_weight[list][i][j] != chroma_def
3018
                           || h->chroma_offset[list][i][j] != 0) {
3019
                            h->use_weight_chroma= 1;
3020
                            h->chroma_weight_flag[list]= 1;
3021
                        }
3022
                    }
3023
                }else{
3024
                    int j;
3025
                    for(j=0; j<2; j++){
3026
                        h->chroma_weight[list][i][j]= chroma_def;
3027
                        h->chroma_offset[list][i][j]= 0;
3028
                    }
3029
                }
3030
            }
3031
        }
3032
        if(h->slice_type_nos != FF_B_TYPE) break;
3033
    }
3034
    h->use_weight= h->use_weight || h->use_weight_chroma;
3035
    return 0;
3036
}
3037

    
3038
static void implicit_weight_table(H264Context *h){
3039
    MpegEncContext * const s = &h->s;
3040
    int ref0, ref1, i;
3041
    int cur_poc = s->current_picture_ptr->poc;
3042

    
3043
    for (i = 0; i < 2; i++) {
3044
        h->luma_weight_flag[i]   = 0;
3045
        h->chroma_weight_flag[i] = 0;
3046
    }
3047

    
3048
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3049
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3050
        h->use_weight= 0;
3051
        h->use_weight_chroma= 0;
3052
        return;
3053
    }
3054

    
3055
    h->use_weight= 2;
3056
    h->use_weight_chroma= 2;
3057
    h->luma_log2_weight_denom= 5;
3058
    h->chroma_log2_weight_denom= 5;
3059

    
3060
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3061
        int poc0 = h->ref_list[0][ref0].poc;
3062
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3063
            int poc1 = h->ref_list[1][ref1].poc;
3064
            int td = av_clip(poc1 - poc0, -128, 127);
3065
            if(td){
3066
                int tb = av_clip(cur_poc - poc0, -128, 127);
3067
                int tx = (16384 + (FFABS(td) >> 1)) / td;
3068
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3069
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3070
                    h->implicit_weight[ref0][ref1] = 32;
3071
                else
3072
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3073
            }else
3074
                h->implicit_weight[ref0][ref1] = 32;
3075
        }
3076
    }
3077
}
3078

    
3079
/**
3080
 * Mark a picture as no longer needed for reference. The refmask
3081
 * argument allows unreferencing of individual fields or the whole frame.
3082
 * If the picture becomes entirely unreferenced, but is being held for
3083
 * display purposes, it is marked as such.
3084
 * @param refmask mask of fields to unreference; the mask is bitwise
3085
 *                anded with the reference marking of pic
3086
 * @return non-zero if pic becomes entirely unreferenced (except possibly
3087
 *         for display purposes) zero if one of the fields remains in
3088
 *         reference
3089
 */
3090
static inline int unreference_pic(H264Context *h, Picture *pic, int refmask){
3091
    int i;
3092
    if (pic->reference &= refmask) {
3093
        return 0;
3094
    } else {
3095
        for(i = 0; h->delayed_pic[i]; i++)
3096
            if(pic == h->delayed_pic[i]){
3097
                pic->reference=DELAYED_PIC_REF;
3098
                break;
3099
            }
3100
        return 1;
3101
    }
3102
}
3103

    
3104
/**
3105
 * instantaneous decoder refresh.
3106
 */
3107
static void idr(H264Context *h){
3108
    int i;
3109

    
3110
    for(i=0; i<16; i++){
3111
        remove_long(h, i, 0);
3112
    }
3113
    assert(h->long_ref_count==0);
3114

    
3115
    for(i=0; i<h->short_ref_count; i++){
3116
        unreference_pic(h, h->short_ref[i], 0);
3117
        h->short_ref[i]= NULL;
3118
    }
3119
    h->short_ref_count=0;
3120
    h->prev_frame_num= 0;
3121
    h->prev_frame_num_offset= 0;
3122
    h->prev_poc_msb=
3123
    h->prev_poc_lsb= 0;
3124
}
3125

    
3126
/* forget old pics after a seek */
3127
static void flush_dpb(AVCodecContext *avctx){
3128
    H264Context *h= avctx->priv_data;
3129
    int i;
3130
    for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
3131
        if(h->delayed_pic[i])
3132
            h->delayed_pic[i]->reference= 0;
3133
        h->delayed_pic[i]= NULL;
3134
    }
3135
    h->outputed_poc= INT_MIN;
3136
    idr(h);
3137
    if(h->s.current_picture_ptr)
3138
        h->s.current_picture_ptr->reference= 0;
3139
    h->s.first_field= 0;
3140
    h->sei_recovery_frame_cnt = -1;
3141
    h->sei_dpb_output_delay = 0;
3142
    h->sei_cpb_removal_delay = -1;
3143
    h->sei_buffering_period_present = 0;
3144
    ff_mpeg_flush(avctx);
3145
}
3146

    
3147
/**
3148
 * Find a Picture in the short term reference list by frame number.
3149
 * @param frame_num frame number to search for
3150
 * @param idx the index into h->short_ref where returned picture is found
3151
 *            undefined if no picture found.
3152
 * @return pointer to the found picture, or NULL if no pic with the provided
3153
 *                 frame number is found
3154
 */
3155
static Picture * find_short(H264Context *h, int frame_num, int *idx){
3156
    MpegEncContext * const s = &h->s;
3157
    int i;
3158

    
3159
    for(i=0; i<h->short_ref_count; i++){
3160
        Picture *pic= h->short_ref[i];
3161
        if(s->avctx->debug&FF_DEBUG_MMCO)
3162
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3163
        if(pic->frame_num == frame_num) {
3164
            *idx = i;
3165
            return pic;
3166
        }
3167
    }
3168
    return NULL;
3169
}
3170

    
3171
/**
3172
 * Remove a picture from the short term reference list by its index in
3173
 * that list.  This does no checking on the provided index; it is assumed
3174
 * to be valid. Other list entries are shifted down.
3175
 * @param i index into h->short_ref of picture to remove.
3176
 */
3177
static void remove_short_at_index(H264Context *h, int i){
3178
    assert(i >= 0 && i < h->short_ref_count);
3179
    h->short_ref[i]= NULL;
3180
    if (--h->short_ref_count)
3181
        memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i)*sizeof(Picture*));
3182
}
3183

    
3184
/**
3185
 *
3186
 * @return the removed picture or NULL if an error occurs
3187
 */
3188
static Picture * remove_short(H264Context *h, int frame_num, int ref_mask){
3189
    MpegEncContext * const s = &h->s;
3190
    Picture *pic;
3191
    int i;
3192

    
3193
    if(s->avctx->debug&FF_DEBUG_MMCO)
3194
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3195

    
3196
    pic = find_short(h, frame_num, &i);
3197
    if (pic){
3198
        if(unreference_pic(h, pic, ref_mask))
3199
        remove_short_at_index(h, i);
3200
    }
3201

    
3202
    return pic;
3203
}
3204

    
3205
/**
3206
 * Remove a picture from the long term reference list by its index in
3207
 * that list.
3208
 * @return the removed picture or NULL if an error occurs
3209
 */
3210
static Picture * remove_long(H264Context *h, int i, int ref_mask){
3211
    Picture *pic;
3212

    
3213
    pic= h->long_ref[i];
3214
    if (pic){
3215
        if(unreference_pic(h, pic, ref_mask)){
3216
            assert(h->long_ref[i]->long_ref == 1);
3217
            h->long_ref[i]->long_ref= 0;
3218
            h->long_ref[i]= NULL;
3219
            h->long_ref_count--;
3220
        }
3221
    }
3222

    
3223
    return pic;
3224
}
3225

    
3226
/**
3227
 * print short term list
3228
 */
3229
static void print_short_term(H264Context *h) {
3230
    uint32_t i;
3231
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3232
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3233
        for(i=0; i<h->short_ref_count; i++){
3234
            Picture *pic= h->short_ref[i];
3235
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3236
        }
3237
    }
3238
}
3239

    
3240
/**
3241
 * print long term list
3242
 */
3243
static void print_long_term(H264Context *h) {
3244
    uint32_t i;
3245
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3246
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3247
        for(i = 0; i < 16; i++){
3248
            Picture *pic= h->long_ref[i];
3249
            if (pic) {
3250
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3251
            }
3252
        }
3253
    }
3254
}
3255

    
3256
/**
3257
 * Executes the reference picture marking (memory management control operations).
3258
 */
3259
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3260
    MpegEncContext * const s = &h->s;
3261
    int i, j;
3262
    int current_ref_assigned=0;
3263
    Picture *av_uninit(pic);
3264

    
3265
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3266
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3267

    
3268
    for(i=0; i<mmco_count; i++){
3269
        int structure, av_uninit(frame_num);
3270
        if(s->avctx->debug&FF_DEBUG_MMCO)
3271
            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);
3272

    
3273
        if(   mmco[i].opcode == MMCO_SHORT2UNUSED
3274
           || mmco[i].opcode == MMCO_SHORT2LONG){
3275
            frame_num = pic_num_extract(h, mmco[i].short_pic_num, &structure);
3276
            pic = find_short(h, frame_num, &j);
3277
            if(!pic){
3278
                if(mmco[i].opcode != MMCO_SHORT2LONG || !h->long_ref[mmco[i].long_arg]
3279
                   || h->long_ref[mmco[i].long_arg]->frame_num != frame_num)
3280
                av_log(h->s.avctx, AV_LOG_ERROR, "mmco: unref short failure\n");
3281
                continue;
3282
            }
3283
        }
3284

    
3285
        switch(mmco[i].opcode){
3286
        case MMCO_SHORT2UNUSED:
3287
            if(s->avctx->debug&FF_DEBUG_MMCO)
3288
                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);
3289
            remove_short(h, frame_num, structure ^ PICT_FRAME);
3290
            break;
3291
        case MMCO_SHORT2LONG:
3292
                if (h->long_ref[mmco[i].long_arg] != pic)
3293
                    remove_long(h, mmco[i].long_arg, 0);
3294

    
3295
                remove_short_at_index(h, j);
3296
                h->long_ref[ mmco[i].long_arg ]= pic;
3297
                if (h->long_ref[ mmco[i].long_arg ]){
3298
                    h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3299
                    h->long_ref_count++;
3300
                }
3301
            break;
3302
        case MMCO_LONG2UNUSED:
3303
            j = pic_num_extract(h, mmco[i].long_arg, &structure);
3304
            pic = h->long_ref[j];
3305
            if (pic) {
3306
                remove_long(h, j, structure ^ PICT_FRAME);
3307
            } else if(s->avctx->debug&FF_DEBUG_MMCO)
3308
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref long failure\n");
3309
            break;
3310
        case MMCO_LONG:
3311
                    // Comment below left from previous code as it is an interresting note.
3312
                    /* First field in pair is in short term list or
3313
                     * at a different long term index.
3314
                     * This is not allowed; see 7.4.3.3, notes 2 and 3.
3315
                     * Report the problem and keep the pair where it is,
3316
                     * and mark this field valid.
3317
                     */
3318

    
3319
            if (h->long_ref[mmco[i].long_arg] != s->current_picture_ptr) {
3320
                remove_long(h, mmco[i].long_arg, 0);
3321

    
3322
                h->long_ref[ mmco[i].long_arg ]= s->current_picture_ptr;
3323
                h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3324
                h->long_ref_count++;
3325
            }
3326

    
3327
            s->current_picture_ptr->reference |= s->picture_structure;
3328
            current_ref_assigned=1;
3329
            break;
3330
        case MMCO_SET_MAX_LONG:
3331
            assert(mmco[i].long_arg <= 16);
3332
            // just remove the long term which index is greater than new max
3333
            for(j = mmco[i].long_arg; j<16; j++){
3334
                remove_long(h, j, 0);
3335
            }
3336
            break;
3337
        case MMCO_RESET:
3338
            while(h->short_ref_count){
3339
                remove_short(h, h->short_ref[0]->frame_num, 0);
3340
            }
3341
            for(j = 0; j < 16; j++) {
3342
                remove_long(h, j, 0);
3343
            }
3344
            s->current_picture_ptr->poc=
3345
            s->current_picture_ptr->field_poc[0]=
3346
            s->current_picture_ptr->field_poc[1]=
3347
            h->poc_lsb=
3348
            h->poc_msb=
3349
            h->frame_num=
3350
            s->current_picture_ptr->frame_num= 0;
3351
            break;
3352
        default: assert(0);
3353
        }
3354
    }
3355

    
3356
    if (!current_ref_assigned) {
3357
        /* Second field of complementary field pair; the first field of
3358
         * which is already referenced. If short referenced, it
3359
         * should be first entry in short_ref. If not, it must exist
3360
         * in long_ref; trying to put it on the short list here is an
3361
         * error in the encoded bit stream (ref: 7.4.3.3, NOTE 2 and 3).
3362
         */
3363
        if (h->short_ref_count && h->short_ref[0] == s->current_picture_ptr) {
3364
            /* Just mark the second field valid */
3365
            s->current_picture_ptr->reference = PICT_FRAME;
3366
        } else if (s->current_picture_ptr->long_ref) {
3367
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term reference "
3368
                                             "assignment for second field "
3369
                                             "in complementary field pair "
3370
                                             "(first field is long term)\n");
3371
        } else {
3372
            pic= remove_short(h, s->current_picture_ptr->frame_num, 0);
3373
            if(pic){
3374
                av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3375
            }
3376

    
3377
            if(h->short_ref_count)
3378
                memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3379

    
3380
            h->short_ref[0]= s->current_picture_ptr;
3381
            h->short_ref_count++;
3382
            s->current_picture_ptr->reference |= s->picture_structure;
3383
        }
3384
    }
3385

    
3386
    if (h->long_ref_count + h->short_ref_count > h->sps.ref_frame_count){
3387

    
3388
        /* We have too many reference frames, probably due to corrupted
3389
         * stream. Need to discard one frame. Prevents overrun of the
3390
         * short_ref and long_ref buffers.
3391
         */
3392
        av_log(h->s.avctx, AV_LOG_ERROR,
3393
               "number of reference frames exceeds max (probably "
3394
               "corrupt input), discarding one\n");
3395

    
3396
        if (h->long_ref_count && !h->short_ref_count) {
3397
            for (i = 0; i < 16; ++i)
3398
                if (h->long_ref[i])
3399
                    break;
3400

    
3401
            assert(i < 16);
3402
            remove_long(h, i, 0);
3403
        } else {
3404
            pic = h->short_ref[h->short_ref_count - 1];
3405
            remove_short(h, pic->frame_num, 0);
3406
        }
3407
    }
3408

    
3409
    print_short_term(h);
3410
    print_long_term(h);
3411
    return 0;
3412
}
3413

    
3414
static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
3415
    MpegEncContext * const s = &h->s;
3416
    int i;
3417

    
3418
    h->mmco_index= 0;
3419
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3420
        s->broken_link= get_bits1(gb) -1;
3421
        if(get_bits1(gb)){
3422
            h->mmco[0].opcode= MMCO_LONG;
3423
            h->mmco[0].long_arg= 0;
3424
            h->mmco_index= 1;
3425
        }
3426
    }else{
3427
        if(get_bits1(gb)){ // adaptive_ref_pic_marking_mode_flag
3428
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
3429
                MMCOOpcode opcode= get_ue_golomb_31(gb);
3430

    
3431
                h->mmco[i].opcode= opcode;
3432
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3433
                    h->mmco[i].short_pic_num= (h->curr_pic_num - get_ue_golomb(gb) - 1) & (h->max_pic_num - 1);
3434
/*                    if(h->mmco[i].short_pic_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_pic_num ] == NULL){
3435
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
3436
                        return -1;
3437
                    }*/
3438
                }
3439
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3440
                    unsigned int long_arg= get_ue_golomb_31(gb);
3441
                    if(long_arg >= 32 || (long_arg >= 16 && !(opcode == MMCO_LONG2UNUSED && FIELD_PICTURE))){
3442
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3443
                        return -1;
3444
                    }
3445
                    h->mmco[i].long_arg= long_arg;
3446
                }
3447

    
3448
                if(opcode > (unsigned)MMCO_LONG){
3449
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3450
                    return -1;
3451
                }
3452
                if(opcode == MMCO_END)
3453
                    break;
3454
            }
3455
            h->mmco_index= i;
3456
        }else{
3457
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3458

    
3459
            if(h->short_ref_count && h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count &&
3460
                    !(FIELD_PICTURE && !s->first_field && s->current_picture_ptr->reference)) {
3461
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3462
                h->mmco[0].short_pic_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3463
                h->mmco_index= 1;
3464
                if (FIELD_PICTURE) {
3465
                    h->mmco[0].short_pic_num *= 2;
3466
                    h->mmco[1].opcode= MMCO_SHORT2UNUSED;
3467
                    h->mmco[1].short_pic_num= h->mmco[0].short_pic_num + 1;
3468
                    h->mmco_index= 2;
3469
                }
3470
            }
3471
        }
3472
    }
3473

    
3474
    return 0;
3475
}
3476

    
3477
static int init_poc(H264Context *h){
3478
    MpegEncContext * const s = &h->s;
3479
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3480
    int field_poc[2];
3481
    Picture *cur = s->current_picture_ptr;
3482

    
3483
    h->frame_num_offset= h->prev_frame_num_offset;
3484
    if(h->frame_num < h->prev_frame_num)
3485
        h->frame_num_offset += max_frame_num;
3486

    
3487
    if(h->sps.poc_type==0){
3488
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3489

    
3490
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3491
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3492
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3493
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3494
        else
3495
            h->poc_msb = h->prev_poc_msb;
3496
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3497
        field_poc[0] =
3498
        field_poc[1] = h->poc_msb + h->poc_lsb;
3499
        if(s->picture_structure == PICT_FRAME)
3500
            field_poc[1] += h->delta_poc_bottom;
3501
    }else if(h->sps.poc_type==1){
3502
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3503
        int i;
3504

    
3505
        if(h->sps.poc_cycle_length != 0)
3506
            abs_frame_num = h->frame_num_offset + h->frame_num;
3507
        else
3508
            abs_frame_num = 0;
3509

    
3510
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
3511
            abs_frame_num--;
3512

    
3513
        expected_delta_per_poc_cycle = 0;
3514
        for(i=0; i < h->sps.poc_cycle_length; i++)
3515
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3516

    
3517
        if(abs_frame_num > 0){
3518
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3519
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3520

    
3521
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3522
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
3523
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3524
        } else
3525
            expectedpoc = 0;
3526

    
3527
        if(h->nal_ref_idc == 0)
3528
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3529

    
3530
        field_poc[0] = expectedpoc + h->delta_poc[0];
3531
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3532

    
3533
        if(s->picture_structure == PICT_FRAME)
3534
            field_poc[1] += h->delta_poc[1];
3535
    }else{
3536
        int poc= 2*(h->frame_num_offset + h->frame_num);
3537

    
3538
        if(!h->nal_ref_idc)
3539
            poc--;
3540

    
3541
        field_poc[0]= poc;
3542
        field_poc[1]= poc;
3543
    }
3544

    
3545
    if(s->picture_structure != PICT_BOTTOM_FIELD)
3546
        s->current_picture_ptr->field_poc[0]= field_poc[0];
3547
    if(s->picture_structure != PICT_TOP_FIELD)
3548
        s->current_picture_ptr->field_poc[1]= field_poc[1];
3549
    cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
3550

    
3551
    return 0;
3552
}
3553

    
3554

    
3555
/**
3556
 * initialize scan tables
3557
 */
3558
static void init_scan_tables(H264Context *h){
3559
    MpegEncContext * const s = &h->s;
3560
    int i;
3561
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
3562
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
3563
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
3564
    }else{
3565
        for(i=0; i<16; i++){
3566
#define T(x) (x>>2) | ((x<<2) & 0xF)
3567
            h->zigzag_scan[i] = T(zigzag_scan[i]);
3568
            h-> field_scan[i] = T( field_scan[i]);
3569
#undef T
3570
        }
3571
    }
3572
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
3573
        memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
3574
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
3575
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
3576
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
3577
    }else{
3578
        for(i=0; i<64; i++){
3579
#define T(x) (x>>3) | ((x&7)<<3)
3580
            h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
3581
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
3582
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
3583
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
3584
#undef T
3585
        }
3586
    }
3587
    if(h->sps.transform_bypass){ //FIXME same ugly
3588
        h->zigzag_scan_q0          = zigzag_scan;
3589
        h->zigzag_scan8x8_q0       = ff_zigzag_direct;
3590
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
3591
        h->field_scan_q0           = field_scan;
3592
        h->field_scan8x8_q0        = field_scan8x8;
3593
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
3594
    }else{
3595
        h->zigzag_scan_q0          = h->zigzag_scan;
3596
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
3597
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
3598
        h->field_scan_q0           = h->field_scan;
3599
        h->field_scan8x8_q0        = h->field_scan8x8;
3600
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
3601
    }
3602
}
3603

    
3604
/**
3605
 * Replicates H264 "master" context to thread contexts.
3606
 */
3607
static void clone_slice(H264Context *dst, H264Context *src)
3608
{
3609
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
3610
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
3611
    dst->s.current_picture      = src->s.current_picture;
3612
    dst->s.linesize             = src->s.linesize;
3613
    dst->s.uvlinesize           = src->s.uvlinesize;
3614
    dst->s.first_field          = src->s.first_field;
3615

    
3616
    dst->prev_poc_msb           = src->prev_poc_msb;
3617
    dst->prev_poc_lsb           = src->prev_poc_lsb;
3618
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
3619
    dst->prev_frame_num         = src->prev_frame_num;
3620
    dst->short_ref_count        = src->short_ref_count;
3621

    
3622
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
3623
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
3624
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
3625
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
3626

    
3627
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
3628
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
3629
}
3630

    
3631
/**
3632
 * decodes a slice header.
3633
 * This will also call MPV_common_init() and frame_start() as needed.
3634
 *
3635
 * @param h h264context
3636
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
3637
 *
3638
 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3639
 */
3640
static int decode_slice_header(H264Context *h, H264Context *h0){
3641
    MpegEncContext * const s = &h->s;
3642
    MpegEncContext * const s0 = &h0->s;
3643
    unsigned int first_mb_in_slice;
3644
    unsigned int pps_id;
3645
    int num_ref_idx_active_override_flag;
3646
    unsigned int slice_type, tmp, i, j;
3647
    int default_ref_list_done = 0;
3648
    int last_pic_structure;
3649

    
3650
    s->dropable= h->nal_ref_idc == 0;
3651

    
3652
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
3653
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
3654
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
3655
    }else{
3656
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
3657
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
3658
    }
3659

    
3660
    first_mb_in_slice= get_ue_golomb(&s->gb);
3661

    
3662
    if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
3663
        h0->current_slice = 0;
3664
        if (!s0->first_field)
3665
            s->current_picture_ptr= NULL;
3666
    }
3667

    
3668
    slice_type= get_ue_golomb_31(&s->gb);
3669
    if(slice_type > 9){
3670
        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);
3671
        return -1;
3672
    }
3673
    if(slice_type > 4){
3674
        slice_type -= 5;
3675
        h->slice_type_fixed=1;
3676
    }else
3677
        h->slice_type_fixed=0;
3678

    
3679
    slice_type= golomb_to_pict_type[ slice_type ];
3680
    if (slice_type == FF_I_TYPE
3681
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
3682
        default_ref_list_done = 1;
3683
    }
3684
    h->slice_type= slice_type;
3685
    h->slice_type_nos= slice_type & 3;
3686

    
3687
    s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
3688
    if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
3689
        av_log(h->s.avctx, AV_LOG_ERROR,
3690
               "B picture before any references, skipping\n");
3691
        return -1;
3692
    }
3693

    
3694
    pps_id= get_ue_golomb(&s->gb);
3695
    if(pps_id>=MAX_PPS_COUNT){
3696
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3697
        return -1;
3698
    }
3699
    if(!h0->pps_buffers[pps_id]) {
3700
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS referenced\n");
3701
        return -1;
3702
    }
3703
    h->pps= *h0->pps_buffers[pps_id];
3704

    
3705
    if(!h0->sps_buffers[h->pps.sps_id]) {
3706
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS referenced\n");
3707
        return -1;
3708
    }
3709
    h->sps = *h0->sps_buffers[h->pps.sps_id];
3710

    
3711
    if(h == h0 && h->dequant_coeff_pps != pps_id){
3712
        h->dequant_coeff_pps = pps_id;
3713
        init_dequant_tables(h);
3714
    }
3715

    
3716
    s->mb_width= h->sps.mb_width;
3717
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3718

    
3719
    h->b_stride=  s->mb_width*4;
3720
    h->b8_stride= s->mb_width*2;
3721

    
3722
    s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
3723
    if(h->sps.frame_mbs_only_flag)
3724
        s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
3725
    else
3726
        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
3727

    
3728
    if (s->context_initialized
3729
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3730
        if(h != h0)
3731
            return -1;   // width / height changed during parallelized decoding
3732
        free_tables(h);
3733
        flush_dpb(s->avctx);
3734
        MPV_common_end(s);
3735
    }
3736
    if (!s->context_initialized) {
3737
        if(h != h0)
3738
            return -1;  // we cant (re-)initialize context during parallel decoding
3739
        if (MPV_common_init(s) < 0)
3740
            return -1;
3741
        s->first_field = 0;
3742

    
3743
        init_scan_tables(h);
3744
        alloc_tables(h);
3745

    
3746
        for(i = 1; i < s->avctx->thread_count; i++) {
3747
            H264Context *c;
3748
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
3749
            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
3750
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
3751
            c->sps = h->sps;
3752
            c->pps = h->pps;
3753
            init_scan_tables(c);
3754
            clone_tables(c, h);
3755
        }
3756

    
3757
        for(i = 0; i < s->avctx->thread_count; i++)
3758
            if(context_init(h->thread_context[i]) < 0)
3759
                return -1;
3760

    
3761
        s->avctx->width = s->width;
3762
        s->avctx->height = s->height;
3763
        s->avctx->sample_aspect_ratio= h->sps.sar;
3764
        if(!s->avctx->sample_aspect_ratio.den)
3765
            s->avctx->sample_aspect_ratio.den = 1;
3766

    
3767
        if(h->sps.timing_info_present_flag){
3768
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
3769
            if(h->x264_build > 0 && h->x264_build < 44)
3770
                s->avctx->time_base.den *= 2;
3771
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3772
                      s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
3773
        }else if(!h->sps.time_scale && !s->avctx->frame_number){
3774
            s->avctx->time_base.den *=2;
3775
            h->sps.time_scale= s->avctx->time_base.den;
3776
        }
3777
    }
3778

    
3779
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3780

    
3781
    h->mb_mbaff = 0;
3782
    h->mb_aff_frame = 0;
3783
    last_pic_structure = s0->picture_structure;
3784
    if(h->sps.frame_mbs_only_flag){
3785
        s->picture_structure= PICT_FRAME;
3786
    }else{
3787
        if(get_bits1(&s->gb)) { //field_pic_flag
3788
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3789
        } else {
3790
            s->picture_structure= PICT_FRAME;
3791
            h->mb_aff_frame = h->sps.mb_aff;
3792
        }
3793
    }
3794
    h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
3795

    
3796
    if(h0->current_slice == 0){
3797
        while(h->frame_num !=  h->prev_frame_num &&
3798
              h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
3799
            av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
3800
            if (frame_start(h) < 0)
3801
                return -1;
3802
            h->prev_frame_num++;
3803
            h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
3804
            s->current_picture_ptr->frame_num= h->prev_frame_num;
3805
            execute_ref_pic_marking(h, NULL, 0);
3806
        }
3807

    
3808
        /* See if we have a decoded first field looking for a pair... */
3809
        if (s0->first_field) {
3810
            assert(s0->current_picture_ptr);
3811
            assert(s0->current_picture_ptr->data[0]);
3812
            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
3813

    
3814
            /* figure out if we have a complementary field pair */
3815
            if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
3816
                /*
3817
                 * Previous field is unmatched. Don't display it, but let it
3818
                 * remain for reference if marked as such.
3819
                 */
3820
                s0->current_picture_ptr = NULL;
3821
                s0->first_field = FIELD_PICTURE;
3822

    
3823
            } else {
3824
                if (h->nal_ref_idc &&
3825
                        s0->current_picture_ptr->reference &&
3826
                        s0->current_picture_ptr->frame_num != h->frame_num) {
3827
                    /*
3828
                     * This and previous field were reference, but had
3829
                     * different frame_nums. Consider this field first in
3830
                     * pair. Throw away previous field except for reference
3831
                     * purposes.
3832
                     */
3833
                    s0->first_field = 1;
3834
                    s0->current_picture_ptr = NULL;
3835

    
3836
                } else {
3837
                    /* Second field in complementary pair */
3838
                    s0->first_field = 0;
3839
                }
3840
            }
3841

    
3842
        } else {
3843
            /* Frame or first field in a potentially complementary pair */
3844
            assert(!s0->current_picture_ptr);
3845
            s0->first_field = FIELD_PICTURE;
3846
        }
3847

    
3848
        if((!FIELD_PICTURE || s0->first_field) && frame_start(h) < 0) {
3849
            s0->first_field = 0;
3850
            return -1;
3851
        }
3852
    }
3853
    if(h != h0)
3854
        clone_slice(h, h0);
3855

    
3856
    s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
3857

    
3858
    assert(s->mb_num == s->mb_width * s->mb_height);
3859
    if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
3860
       first_mb_in_slice                    >= s->mb_num){
3861
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3862
        return -1;
3863
    }
3864
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3865
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
3866
    if (s->picture_structure == PICT_BOTTOM_FIELD)
3867
        s->resync_mb_y = s->mb_y = s->mb_y + 1;
3868
    assert(s->mb_y < s->mb_height);
3869

    
3870
    if(s->picture_structure==PICT_FRAME){
3871
        h->curr_pic_num=   h->frame_num;
3872
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3873
    }else{
3874
        h->curr_pic_num= 2*h->frame_num + 1;
3875
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3876
    }
3877

    
3878
    if(h->nal_unit_type == NAL_IDR_SLICE){
3879
        get_ue_golomb(&s->gb); /* idr_pic_id */
3880
    }
3881

    
3882
    if(h->sps.poc_type==0){
3883
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3884

    
3885
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3886
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3887
        }
3888
    }
3889

    
3890
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3891
        h->delta_poc[0]= get_se_golomb(&s->gb);
3892

    
3893
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3894
            h->delta_poc[1]= get_se_golomb(&s->gb);
3895
    }
3896

    
3897
    init_poc(h);
3898

    
3899
    if(h->pps.redundant_pic_cnt_present){
3900
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3901
    }
3902

    
3903
    //set defaults, might be overridden a few lines later
3904
    h->ref_count[0]= h->pps.ref_count[0];
3905
    h->ref_count[1]= h->pps.ref_count[1];
3906

    
3907
    if(h->slice_type_nos != FF_I_TYPE){
3908
        if(h->slice_type_nos == FF_B_TYPE){
3909
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3910
        }
3911
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3912

    
3913
        if(num_ref_idx_active_override_flag){
3914
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3915
            if(h->slice_type_nos==FF_B_TYPE)
3916
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3917

    
3918
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
3919
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3920
                h->ref_count[0]= h->ref_count[1]= 1;
3921
                return -1;
3922
            }
3923
        }
3924
        if(h->slice_type_nos == FF_B_TYPE)
3925
            h->list_count= 2;
3926
        else
3927
            h->list_count= 1;
3928
    }else
3929
        h->list_count= 0;
3930

    
3931
    if(!default_ref_list_done){
3932
        fill_default_ref_list(h);
3933
    }
3934

    
3935
    if(h->slice_type_nos!=FF_I_TYPE && decode_ref_pic_list_reordering(h) < 0)
3936
        return -1;
3937

    
3938
    if(h->slice_type_nos!=FF_I_TYPE){
3939
        s->last_picture_ptr= &h->ref_list[0][0];
3940
        ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3941
    }
3942
    if(h->slice_type_nos==FF_B_TYPE){
3943
        s->next_picture_ptr= &h->ref_list[1][0];
3944
        ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3945
    }
3946

    
3947
    if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
3948
       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
3949
        pred_weight_table(h);
3950
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
3951
        implicit_weight_table(h);
3952
    else {
3953
        h->use_weight = 0;
3954
        for (i = 0; i < 2; i++) {
3955
            h->luma_weight_flag[i]   = 0;
3956
            h->chroma_weight_flag[i] = 0;
3957
        }
3958
    }
3959

    
3960
    if(h->nal_ref_idc)
3961
        decode_ref_pic_marking(h0, &s->gb);
3962

    
3963
    if(FRAME_MBAFF)
3964
        fill_mbaff_ref_list(h);
3965

    
3966
    if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
3967
        direct_dist_scale_factor(h);
3968
    direct_ref_list_init(h);
3969

    
3970
    if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
3971
        tmp = get_ue_golomb_31(&s->gb);
3972
        if(tmp > 2){
3973
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3974
            return -1;
3975
        }
3976
        h->cabac_init_idc= tmp;
3977
    }
3978

    
3979
    h->last_qscale_diff = 0;
3980
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3981
    if(tmp>51){
3982
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3983
        return -1;
3984
    }
3985
    s->qscale= tmp;
3986
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3987
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3988
    //FIXME qscale / qp ... stuff
3989
    if(h->slice_type == FF_SP_TYPE){
3990
        get_bits1(&s->gb); /* sp_for_switch_flag */
3991
    }
3992
    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
3993
        get_se_golomb(&s->gb); /* slice_qs_delta */
3994
    }
3995

    
3996
    h->deblocking_filter = 1;
3997
    h->slice_alpha_c0_offset = 0;
3998
    h->slice_beta_offset = 0;
3999
    if( h->pps.deblocking_filter_parameters_present ) {
4000
        tmp= get_ue_golomb_31(&s->gb);
4001
        if(tmp > 2){
4002
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
4003
            return -1;
4004
        }
4005
        h->deblocking_filter= tmp;
4006
        if(h->deblocking_filter < 2)
4007
            h->deblocking_filter^= 1; // 1<->0
4008

    
4009
        if( h->deblocking_filter ) {
4010
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
4011
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
4012
        }
4013
    }
4014

    
4015
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
4016
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
4017
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
4018
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
4019
        h->deblocking_filter= 0;
4020

    
4021
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
4022
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
4023
            /* Cheat slightly for speed:
4024
               Do not bother to deblock across slices. */
4025
            h->deblocking_filter = 2;
4026
        } else {
4027
            h0->max_contexts = 1;
4028
            if(!h0->single_decode_warning) {
4029
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
4030
                h0->single_decode_warning = 1;
4031
            }
4032
            if(h != h0)
4033
                return 1; // deblocking switched inside frame
4034
        }
4035
    }
4036

    
4037
#if 0 //FMO
4038
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
4039
        slice_group_change_cycle= get_bits(&s->gb, ?);
4040
#endif
4041

    
4042
    h0->last_slice_type = slice_type;
4043
    h->slice_num = ++h0->current_slice;
4044
    if(h->slice_num >= MAX_SLICES){
4045
        av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
4046
    }
4047

    
4048
    for(j=0; j<2; j++){
4049
        int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
4050
        ref2frm[0]=
4051
        ref2frm[1]= -1;
4052
        for(i=0; i<16; i++)
4053
            ref2frm[i+2]= 4*h->ref_list[j][i].frame_num
4054
                          +(h->ref_list[j][i].reference&3);
4055
        ref2frm[18+0]=
4056
        ref2frm[18+1]= -1;
4057
        for(i=16; i<48; i++)
4058
            ref2frm[i+4]= 4*h->ref_list[j][i].frame_num
4059
                          +(h->ref_list[j][i].reference&3);
4060
    }
4061

    
4062
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
4063
    h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
4064

    
4065
    s->avctx->refs= h->sps.ref_frame_count;
4066

    
4067
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4068
        av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
4069
               h->slice_num,
4070
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
4071
               first_mb_in_slice,
4072
               av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
4073
               pps_id, h->frame_num,
4074
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4075
               h->ref_count[0], h->ref_count[1],
4076
               s->qscale,
4077
               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4078
               h->use_weight,
4079
               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
4080
               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
4081
               );
4082
    }
4083

    
4084
    return 0;
4085
}
4086

    
4087
/**
4088
 *
4089
 */
4090
static inline int get_level_prefix(GetBitContext *gb){
4091
    unsigned int buf;
4092
    int log;
4093

    
4094
    OPEN_READER(re, gb);
4095
    UPDATE_CACHE(re, gb);
4096
    buf=GET_CACHE(re, gb);
4097

    
4098
    log= 32 - av_log2(buf);
4099
#ifdef TRACE
4100
    print_bin(buf>>(32-log), log);
4101
    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__);
4102
#endif
4103

    
4104
    LAST_SKIP_BITS(re, gb, log);
4105
    CLOSE_READER(re, gb);
4106

    
4107
    return log-1;
4108
}
4109

    
4110
static inline int get_dct8x8_allowed(H264Context *h){
4111
    if(h->sps.direct_8x8_inference_flag)
4112
        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
4113
    else
4114
        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
4115
}
4116

    
4117
/**
4118
 * decodes a residual block.
4119
 * @param n block index
4120
 * @param scantable scantable
4121
 * @param max_coeff number of coefficients in the block
4122
 * @return <0 if an error occurred
4123
 */
4124
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
4125
    MpegEncContext * const s = &h->s;
4126
    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};
4127
    int level[16];
4128
    int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;
4129

    
4130
    //FIXME put trailing_onex into the context
4131

    
4132
    if(n == CHROMA_DC_BLOCK_INDEX){
4133
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4134
        total_coeff= coeff_token>>2;
4135
    }else{
4136
        if(n == LUMA_DC_BLOCK_INDEX){
4137
            total_coeff= pred_non_zero_count(h, 0);
4138
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4139
            total_coeff= coeff_token>>2;
4140
        }else{
4141
            total_coeff= pred_non_zero_count(h, n);
4142
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4143
            total_coeff= coeff_token>>2;
4144
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4145
        }
4146
    }
4147

    
4148
    //FIXME set last_non_zero?
4149

    
4150
    if(total_coeff==0)
4151
        return 0;
4152
    if(total_coeff > (unsigned)max_coeff) {
4153
        av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
4154
        return -1;
4155
    }
4156

    
4157
    trailing_ones= coeff_token&3;
4158
    tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
4159
    assert(total_coeff<=16);
4160

    
4161
    i = show_bits(gb, 3);
4162
    skip_bits(gb, trailing_ones);
4163
    level[0] = 1-((i&4)>>1);
4164
    level[1] = 1-((i&2)   );
4165
    level[2] = 1-((i&1)<<1);
4166

    
4167
    if(trailing_ones<total_coeff) {
4168
        int mask, prefix;
4169
        int suffix_length = total_coeff > 10 && trailing_ones < 3;
4170
        int bitsi= show_bits(gb, LEVEL_TAB_BITS);
4171
        int level_code= cavlc_level_tab[suffix_length][bitsi][0];
4172

    
4173
        skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
4174
        if(level_code >= 100){
4175
            prefix= level_code - 100;
4176
            if(prefix == LEVEL_TAB_BITS)
4177
                prefix += get_level_prefix(gb);
4178

    
4179
            //first coefficient has suffix_length equal to 0 or 1
4180
            if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4181
                if(suffix_length)
4182
                    level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4183
                else
4184
                    level_code= (prefix<<suffix_length); //part
4185
            }else if(prefix==14){
4186
                if(suffix_length)
4187
                    level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4188
                else
4189
                    level_code= prefix + get_bits(gb, 4); //part
4190
            }else{
4191
                level_code= (15<<suffix_length) + get_bits(gb, prefix-3); //part
4192
                if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
4193
                if(prefix>=16)
4194
                    level_code += (1<<(prefix-3))-4096;
4195
            }
4196

    
4197
            if(trailing_ones < 3) level_code += 2;
4198

    
4199
            suffix_length = 2;
4200
            mask= -(level_code&1);
4201
            level[trailing_ones]= (((2+level_code)>>1) ^ mask) - mask;
4202
        }else{
4203
            if(trailing_ones < 3) level_code += (level_code>>31)|1;
4204

    
4205
            suffix_length = 1;
4206
            if(level_code + 3U > 6U)
4207
                suffix_length++;
4208
            level[trailing_ones]= level_code;
4209
        }
4210

    
4211
        //remaining coefficients have suffix_length > 0
4212
        for(i=trailing_ones+1;i<total_coeff;i++) {
4213
            static const unsigned int suffix_limit[7] = {0,3,6,12,24,48,INT_MAX };
4214
            int bitsi= show_bits(gb, LEVEL_TAB_BITS);
4215
            level_code= cavlc_level_tab[suffix_length][bitsi][0];
4216

    
4217
            skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
4218
            if(level_code >= 100){
4219
                prefix= level_code - 100;
4220
                if(prefix == LEVEL_TAB_BITS){
4221
                    prefix += get_level_prefix(gb);
4222
                }
4223
                if(prefix<15){
4224
                    level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
4225
                }else{
4226
                    level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
4227
                    if(prefix>=16)
4228
                        level_code += (1<<(prefix-3))-4096;
4229
                }
4230
                mask= -(level_code&1);
4231
                level_code= (((2+level_code)>>1) ^ mask) - mask;
4232
            }
4233
            level[i]= level_code;
4234

    
4235
            if(suffix_limit[suffix_length] + level_code > 2U*suffix_limit[suffix_length])
4236
                suffix_length++;
4237
        }
4238
    }
4239

    
4240
    if(total_coeff == max_coeff)
4241
        zeros_left=0;
4242
    else{
4243
        if(n == CHROMA_DC_BLOCK_INDEX)
4244
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4245
        else
4246
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4247
    }
4248

    
4249
    coeff_num = zeros_left + total_coeff - 1;
4250
    j = scantable[coeff_num];
4251
    if(n > 24){
4252
        block[j] = level[0];
4253
        for(i=1;i<total_coeff;i++) {
4254
            if(zeros_left <= 0)
4255
                run_before = 0;
4256
            else if(zeros_left < 7){
4257
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4258
            }else{
4259
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4260
            }
4261
            zeros_left -= run_before;
4262
            coeff_num -= 1 + run_before;
4263
            j= scantable[ coeff_num ];
4264

    
4265
            block[j]= level[i];
4266
        }
4267
    }else{
4268
        block[j] = (level[0] * qmul[j] + 32)>>6;
4269
        for(i=1;i<total_coeff;i++) {
4270
            if(zeros_left <= 0)
4271
                run_before = 0;
4272
            else if(zeros_left < 7){
4273
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4274
            }else{
4275
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4276
            }
4277
            zeros_left -= run_before;
4278
            coeff_num -= 1 + run_before;
4279
            j= scantable[ coeff_num ];
4280

    
4281
            block[j]= (level[i] * qmul[j] + 32)>>6;
4282
        }
4283
    }
4284

    
4285
    if(zeros_left<0){
4286
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4287
        return -1;
4288
    }
4289

    
4290
    return 0;
4291
}
4292

    
4293
static void predict_field_decoding_flag(H264Context *h){
4294
    MpegEncContext * const s = &h->s;
4295
    const int mb_xy= h->mb_xy;
4296
    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
4297
                ? s->current_picture.mb_type[mb_xy-1]
4298
                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
4299
                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
4300
                : 0;
4301
    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4302
}
4303

    
4304
/**
4305
 * decodes a P_SKIP or B_SKIP macroblock
4306
 */
4307
static void decode_mb_skip(H264Context *h){
4308
    MpegEncContext * const s = &h->s;
4309
    const int mb_xy= h->mb_xy;
4310
    int mb_type=0;
4311

    
4312
    memset(h->non_zero_count[mb_xy], 0, 16);
4313
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4314

    
4315
    if(MB_FIELD)
4316
        mb_type|= MB_TYPE_INTERLACED;
4317

    
4318
    if( h->slice_type_nos == FF_B_TYPE )
4319
    {
4320
        // just for fill_caches. pred_direct_motion will set the real mb_type
4321
        mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4322

    
4323
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4324
        pred_direct_motion(h, &mb_type);
4325
        mb_type|= MB_TYPE_SKIP;
4326
    }
4327
    else
4328
    {
4329
        int mx, my;
4330
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4331

    
4332
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4333
        pred_pskip_motion(h, &mx, &my);
4334
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4335
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4336
    }
4337

    
4338
    write_back_motion(h, mb_type);
4339
    s->current_picture.mb_type[mb_xy]= mb_type;
4340
    s->current_picture.qscale_table[mb_xy]= s->qscale;
4341
    h->slice_table[ mb_xy ]= h->slice_num;
4342
    h->prev_mb_skipped= 1;
4343
}
4344

    
4345
/**
4346
 * decodes a macroblock
4347
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4348
 */
4349
static int decode_mb_cavlc(H264Context *h){
4350
    MpegEncContext * const s = &h->s;
4351
    int mb_xy;
4352
    int partition_count;
4353
    unsigned int mb_type, cbp;
4354
    int dct8x8_allowed= h->pps.transform_8x8_mode;
4355

    
4356
    mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4357

    
4358
    tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4359
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4360
                down the code */
4361
    if(h->slice_type_nos != FF_I_TYPE){
4362
        if(s->mb_skip_run==-1)
4363
            s->mb_skip_run= get_ue_golomb(&s->gb);
4364

    
4365
        if (s->mb_skip_run--) {
4366
            if(FRAME_MBAFF && (s->mb_y&1) == 0){
4367
                if(s->mb_skip_run==0)
4368
                    h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4369
                else
4370
                    predict_field_decoding_flag(h);
4371
            }
4372
            decode_mb_skip(h);
4373
            return 0;
4374
        }
4375
    }
4376
    if(FRAME_MBAFF){
4377
        if( (s->mb_y&1) == 0 )
4378
            h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4379
    }
4380

    
4381
    h->prev_mb_skipped= 0;
4382

    
4383
    mb_type= get_ue_golomb(&s->gb);
4384
    if(h->slice_type_nos == FF_B_TYPE){
4385
        if(mb_type < 23){
4386
            partition_count= b_mb_type_info[mb_type].partition_count;
4387
            mb_type=         b_mb_type_info[mb_type].type;
4388
        }else{
4389
            mb_type -= 23;
4390
            goto decode_intra_mb;
4391
        }
4392
    }else if(h->slice_type_nos == FF_P_TYPE){
4393
        if(mb_type < 5){
4394
            partition_count= p_mb_type_info[mb_type].partition_count;
4395
            mb_type=         p_mb_type_info[mb_type].type;
4396
        }else{
4397
            mb_type -= 5;
4398
            goto decode_intra_mb;
4399
        }
4400
    }else{
4401
       assert(h->slice_type_nos == FF_I_TYPE);
4402
        if(h->slice_type == FF_SI_TYPE && mb_type)
4403
            mb_type--;
4404
decode_intra_mb:
4405
        if(mb_type > 25){
4406
            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);
4407
            return -1;
4408
        }
4409
        partition_count=0;
4410
        cbp= i_mb_type_info[mb_type].cbp;
4411
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4412
        mb_type= i_mb_type_info[mb_type].type;
4413
    }
4414

    
4415
    if(MB_FIELD)
4416
        mb_type |= MB_TYPE_INTERLACED;
4417

    
4418
    h->slice_table[ mb_xy ]= h->slice_num;
4419

    
4420
    if(IS_INTRA_PCM(mb_type)){
4421
        unsigned int x;
4422

    
4423
        // We assume these blocks are very rare so we do not optimize it.
4424
        align_get_bits(&s->gb);
4425

    
4426
        // The pixels are stored in the same order as levels in h->mb array.
4427
        for(x=0; x < (CHROMA ? 384 : 256); x++){
4428
            ((uint8_t*)h->mb)[x]= get_bits(&s->gb, 8);
4429
        }
4430

    
4431
        // In deblocking, the quantizer is 0
4432
        s->current_picture.qscale_table[mb_xy]= 0;
4433
        // All coeffs are present
4434
        memset(h->non_zero_count[mb_xy], 16, 16);
4435

    
4436
        s->current_picture.mb_type[mb_xy]= mb_type;
4437
        return 0;
4438
    }
4439

    
4440
    if(MB_MBAFF){
4441
        h->ref_count[0] <<= 1;
4442
        h->ref_count[1] <<= 1;
4443
    }
4444

    
4445
    fill_caches(h, mb_type, 0);
4446

    
4447
    //mb_pred
4448
    if(IS_INTRA(mb_type)){
4449
        int pred_mode;
4450
//            init_top_left_availability(h);
4451
        if(IS_INTRA4x4(mb_type)){
4452
            int i;
4453
            int di = 1;
4454
            if(dct8x8_allowed && get_bits1(&s->gb)){
4455
                mb_type |= MB_TYPE_8x8DCT;
4456
                di = 4;
4457
            }
4458

    
4459
//                fill_intra4x4_pred_table(h);
4460
            for(i=0; i<16; i+=di){
4461
                int mode= pred_intra_mode(h, i);
4462

    
4463
                if(!get_bits1(&s->gb)){
4464
                    const int rem_mode= get_bits(&s->gb, 3);
4465
                    mode = rem_mode + (rem_mode >= mode);
4466
                }
4467

    
4468
                if</