Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 79f4494a

History | View | Annotate | Download (313 KB)

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

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

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

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

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

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

    
51
static VLC coeff_token_vlc[4];
52
static VLC chroma_dc_coeff_token_vlc;
53

    
54
static VLC total_zeros_vlc[15];
55
static VLC chroma_dc_total_zeros_vlc[3];
56

    
57
static VLC run_vlc[6];
58
static VLC run7_vlc;
59

    
60
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
61
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
62
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);
63
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);
64

    
65
static av_always_inline uint32_t pack16to32(int a, int b){
66
#ifdef WORDS_BIGENDIAN
67
   return (b&0xFFFF) + (a<<16);
68
#else
69
   return (a&0xFFFF) + (b<<16);
70
#endif
71
}
72

    
73
const uint8_t ff_rem6[52]={
74
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,
75
};
76

    
77
const uint8_t ff_div6[52]={
78
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,
79
};
80

    
81

    
82
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
83
    MpegEncContext * const s = &h->s;
84
    const int mb_xy= h->mb_xy;
85
    int topleft_xy, top_xy, topright_xy, left_xy[2];
86
    int topleft_type, top_type, topright_type, left_type[2];
87
    int left_block[8];
88
    int topleft_partition= -1;
89
    int i;
90

    
91
    top_xy     = mb_xy  - (s->mb_stride << FIELD_PICTURE);
92

    
93
    //FIXME deblocking could skip the intra and nnz parts.
94
    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
95
        return;
96

    
97
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
98
     * stuff, I can't imagine that these complex rules are worth it. */
99

    
100
    topleft_xy = top_xy - 1;
101
    topright_xy= top_xy + 1;
102
    left_xy[1] = left_xy[0] = mb_xy-1;
103
    left_block[0]= 0;
104
    left_block[1]= 1;
105
    left_block[2]= 2;
106
    left_block[3]= 3;
107
    left_block[4]= 7;
108
    left_block[5]= 10;
109
    left_block[6]= 8;
110
    left_block[7]= 11;
111
    if(FRAME_MBAFF){
112
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
113
        const int top_pair_xy      = pair_xy     - s->mb_stride;
114
        const int topleft_pair_xy  = top_pair_xy - 1;
115
        const int topright_pair_xy = top_pair_xy + 1;
116
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
117
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
118
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
119
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
120
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
121
        const int bottom = (s->mb_y & 1);
122
        tprintf(s->avctx, "fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
123
        if (bottom
124
                ? !curr_mb_frame_flag // bottom macroblock
125
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
126
                ) {
127
            top_xy -= s->mb_stride;
128
        }
129
        if (bottom
130
                ? !curr_mb_frame_flag // bottom macroblock
131
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
132
                ) {
133
            topleft_xy -= s->mb_stride;
134
        } else if(bottom && curr_mb_frame_flag && !left_mb_frame_flag) {
135
            topleft_xy += s->mb_stride;
136
            // take topleft mv from the middle of the mb, as opposed to all other modes which use the bottom-right partition
137
            topleft_partition = 0;
138
        }
139
        if (bottom
140
                ? !curr_mb_frame_flag // bottom macroblock
141
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
142
                ) {
143
            topright_xy -= s->mb_stride;
144
        }
145
        if (left_mb_frame_flag != curr_mb_frame_flag) {
146
            left_xy[1] = left_xy[0] = pair_xy - 1;
147
            if (curr_mb_frame_flag) {
148
                if (bottom) {
149
                    left_block[0]= 2;
150
                    left_block[1]= 2;
151
                    left_block[2]= 3;
152
                    left_block[3]= 3;
153
                    left_block[4]= 8;
154
                    left_block[5]= 11;
155
                    left_block[6]= 8;
156
                    left_block[7]= 11;
157
                } else {
158
                    left_block[0]= 0;
159
                    left_block[1]= 0;
160
                    left_block[2]= 1;
161
                    left_block[3]= 1;
162
                    left_block[4]= 7;
163
                    left_block[5]= 10;
164
                    left_block[6]= 7;
165
                    left_block[7]= 10;
166
                }
167
            } else {
168
                left_xy[1] += s->mb_stride;
169
                //left_block[0]= 0;
170
                left_block[1]= 2;
171
                left_block[2]= 0;
172
                left_block[3]= 2;
173
                //left_block[4]= 7;
174
                left_block[5]= 10;
175
                left_block[6]= 7;
176
                left_block[7]= 10;
177
            }
178
        }
179
    }
180

    
181
    h->top_mb_xy = top_xy;
182
    h->left_mb_xy[0] = left_xy[0];
183
    h->left_mb_xy[1] = left_xy[1];
184
    if(for_deblock){
185
        topleft_type = 0;
186
        topright_type = 0;
187
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
188
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
189
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
190

    
191
        if(FRAME_MBAFF && !IS_INTRA(mb_type)){
192
            int list;
193
            int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
194
            for(i=0; i<16; i++)
195
                h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
196
            for(list=0; list<h->list_count; list++){
197
                if(USES_LIST(mb_type,list)){
198
                    uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
199
                    uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
200
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
201
                    for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
202
                        dst[0] = src[0];
203
                        dst[1] = src[1];
204
                        dst[2] = src[2];
205
                        dst[3] = src[3];
206
                    }
207
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
208
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
209
                    ref += h->b8_stride;
210
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
211
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
212
                }else{
213
                    fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
214
                    fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
215
                }
216
            }
217
        }
218
    }else{
219
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
220
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
221
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
222
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
223
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
224
    }
225

    
226
    if(IS_INTRA(mb_type)){
227
        h->topleft_samples_available=
228
        h->top_samples_available=
229
        h->left_samples_available= 0xFFFF;
230
        h->topright_samples_available= 0xEEEA;
231

    
232
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
233
            h->topleft_samples_available= 0xB3FF;
234
            h->top_samples_available= 0x33FF;
235
            h->topright_samples_available= 0x26EA;
236
        }
237
        for(i=0; i<2; i++){
238
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
239
                h->topleft_samples_available&= 0xDF5F;
240
                h->left_samples_available&= 0x5F5F;
241
            }
242
        }
243

    
244
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
245
            h->topleft_samples_available&= 0x7FFF;
246

    
247
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
248
            h->topright_samples_available&= 0xFBFF;
249

    
250
        if(IS_INTRA4x4(mb_type)){
251
            if(IS_INTRA4x4(top_type)){
252
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
253
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
254
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
255
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
256
            }else{
257
                int pred;
258
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
259
                    pred= -1;
260
                else{
261
                    pred= 2;
262
                }
263
                h->intra4x4_pred_mode_cache[4+8*0]=
264
                h->intra4x4_pred_mode_cache[5+8*0]=
265
                h->intra4x4_pred_mode_cache[6+8*0]=
266
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
267
            }
268
            for(i=0; i<2; i++){
269
                if(IS_INTRA4x4(left_type[i])){
270
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
271
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
272
                }else{
273
                    int pred;
274
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
275
                        pred= -1;
276
                    else{
277
                        pred= 2;
278
                    }
279
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
280
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
281
                }
282
            }
283
        }
284
    }
285

    
286

    
287
/*
288
0 . T T. T T T T
289
1 L . .L . . . .
290
2 L . .L . . . .
291
3 . T TL . . . .
292
4 L . .L . . . .
293
5 L . .. . . . .
294
*/
295
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
296
    if(top_type){
297
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
298
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
299
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
300
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
301

    
302
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
303
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
304

    
305
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
306
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
307

    
308
    }else{
309
        h->non_zero_count_cache[4+8*0]=
310
        h->non_zero_count_cache[5+8*0]=
311
        h->non_zero_count_cache[6+8*0]=
312
        h->non_zero_count_cache[7+8*0]=
313

    
314
        h->non_zero_count_cache[1+8*0]=
315
        h->non_zero_count_cache[2+8*0]=
316

    
317
        h->non_zero_count_cache[1+8*3]=
318
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
319

    
320
    }
321

    
322
    for (i=0; i<2; i++) {
323
        if(left_type[i]){
324
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
325
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
326
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
327
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
328
        }else{
329
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
330
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
331
            h->non_zero_count_cache[0+8*1 +   8*i]=
332
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
333
        }
334
    }
335

    
336
    if( h->pps.cabac ) {
337
        // top_cbp
338
        if(top_type) {
339
            h->top_cbp = h->cbp_table[top_xy];
340
        } else if(IS_INTRA(mb_type)) {
341
            h->top_cbp = 0x1C0;
342
        } else {
343
            h->top_cbp = 0;
344
        }
345
        // left_cbp
346
        if (left_type[0]) {
347
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
348
        } else if(IS_INTRA(mb_type)) {
349
            h->left_cbp = 0x1C0;
350
        } else {
351
            h->left_cbp = 0;
352
        }
353
        if (left_type[0]) {
354
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
355
        }
356
        if (left_type[1]) {
357
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
358
        }
359
    }
360

    
361
#if 1
362
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
363
        int list;
364
        for(list=0; list<h->list_count; list++){
365
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
366
                /*if(!h->mv_cache_clean[list]){
367
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
368
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
369
                    h->mv_cache_clean[list]= 1;
370
                }*/
371
                continue;
372
            }
373
            h->mv_cache_clean[list]= 0;
374

    
375
            if(USES_LIST(top_type, list)){
376
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
377
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
378
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
379
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
380
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
381
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
382
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
383
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
384
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
385
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
386
            }else{
387
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
388
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
389
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
390
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
391
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
392
            }
393

    
394
            for(i=0; i<2; i++){
395
                int cache_idx = scan8[0] - 1 + i*2*8;
396
                if(USES_LIST(left_type[i], list)){
397
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
398
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
399
                    *(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]];
400
                    *(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]];
401
                    h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
402
                    h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
403
                }else{
404
                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
405
                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
406
                    h->ref_cache[list][cache_idx  ]=
407
                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
408
                }
409
            }
410

    
411
            if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
412
                continue;
413

    
414
            if(USES_LIST(topleft_type, list)){
415
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
416
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
417
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
418
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
419
            }else{
420
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
421
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
422
            }
423

    
424
            if(USES_LIST(topright_type, list)){
425
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
426
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
427
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
428
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
429
            }else{
430
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
431
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
432
            }
433

    
434
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
435
                continue;
436

    
437
            h->ref_cache[list][scan8[5 ]+1] =
438
            h->ref_cache[list][scan8[7 ]+1] =
439
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
440
            h->ref_cache[list][scan8[4 ]] =
441
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
442
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
443
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
444
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
445
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
446
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
447

    
448
            if( h->pps.cabac ) {
449
                /* XXX beurk, Load mvd */
450
                if(USES_LIST(top_type, list)){
451
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
452
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
453
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
454
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
455
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
456
                }else{
457
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
458
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
459
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
460
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
461
                }
462
                if(USES_LIST(left_type[0], list)){
463
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
464
                    *(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]];
465
                    *(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]];
466
                }else{
467
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
468
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
469
                }
470
                if(USES_LIST(left_type[1], list)){
471
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
472
                    *(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]];
473
                    *(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]];
474
                }else{
475
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
476
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
477
                }
478
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
479
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
480
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
481
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
482
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
483

    
484
                if(h->slice_type == FF_B_TYPE){
485
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
486

    
487
                    if(IS_DIRECT(top_type)){
488
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
489
                    }else if(IS_8X8(top_type)){
490
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
491
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
492
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
493
                    }else{
494
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
495
                    }
496

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

    
504
                    if(IS_DIRECT(left_type[1]))
505
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
506
                    else if(IS_8X8(left_type[1]))
507
                        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)];
508
                    else
509
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
510
                }
511
            }
512

    
513
            if(FRAME_MBAFF){
514
#define MAP_MVS\
515
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
516
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
517
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
518
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
519
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
520
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
521
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
522
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
523
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
524
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
525
                if(MB_FIELD){
526
#define MAP_F2F(idx, mb_type)\
527
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
528
                        h->ref_cache[list][idx] <<= 1;\
529
                        h->mv_cache[list][idx][1] /= 2;\
530
                        h->mvd_cache[list][idx][1] /= 2;\
531
                    }
532
                    MAP_MVS
533
#undef MAP_F2F
534
                }else{
535
#define MAP_F2F(idx, mb_type)\
536
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
537
                        h->ref_cache[list][idx] >>= 1;\
538
                        h->mv_cache[list][idx][1] <<= 1;\
539
                        h->mvd_cache[list][idx][1] <<= 1;\
540
                    }
541
                    MAP_MVS
542
#undef MAP_F2F
543
                }
544
            }
545
        }
546
    }
547
#endif
548

    
549
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
550
}
551

    
552
static inline void write_back_intra_pred_mode(H264Context *h){
553
    const int mb_xy= h->mb_xy;
554

    
555
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
556
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
557
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
558
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
559
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
560
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
561
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
562
}
563

    
564
/**
565
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
566
 */
567
static inline int check_intra4x4_pred_mode(H264Context *h){
568
    MpegEncContext * const s = &h->s;
569
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
570
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
571
    int i;
572

    
573
    if(!(h->top_samples_available&0x8000)){
574
        for(i=0; i<4; i++){
575
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
576
            if(status<0){
577
                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);
578
                return -1;
579
            } else if(status){
580
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
581
            }
582
        }
583
    }
584

    
585
    if(!(h->left_samples_available&0x8000)){
586
        for(i=0; i<4; i++){
587
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
588
            if(status<0){
589
                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);
590
                return -1;
591
            } else if(status){
592
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
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&0x8000)){
622
        mode= left[ mode ];
623
        if(mode<0){
624
            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);
625
            return -1;
626
        }
627
    }
628

    
629
    return mode;
630
}
631

    
632
/**
633
 * gets the predicted intra4x4 prediction mode.
634
 */
635
static inline int pred_intra_mode(H264Context *h, int n){
636
    const int index8= scan8[n];
637
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
638
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
639
    const int min= FFMIN(left, top);
640

    
641
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
642

    
643
    if(min<0) return DC_PRED;
644
    else      return min;
645
}
646

    
647
static inline void write_back_non_zero_count(H264Context *h){
648
    const int mb_xy= h->mb_xy;
649

    
650
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
651
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
652
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
653
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
654
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
655
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
656
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
657

    
658
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
659
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
660
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
661

    
662
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
663
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
664
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
665

    
666
    if(FRAME_MBAFF){
667
        // store all luma nnzs, for deblocking
668
        int v = 0, i;
669
        for(i=0; i<16; i++)
670
            v += (!!h->non_zero_count_cache[scan8[i]]) << i;
671
        *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
672
    }
673
}
674

    
675
/**
676
 * gets the predicted number of non zero coefficients.
677
 * @param n block index
678
 */
679
static inline int pred_non_zero_count(H264Context *h, int n){
680
    const int index8= scan8[n];
681
    const int left= h->non_zero_count_cache[index8 - 1];
682
    const int top = h->non_zero_count_cache[index8 - 8];
683
    int i= left + top;
684

    
685
    if(i<64) i= (i+1)>>1;
686

    
687
    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
688

    
689
    return i&31;
690
}
691

    
692
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
693
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
694
    MpegEncContext *s = &h->s;
695

    
696
    /* there is no consistent mapping of mvs to neighboring locations that will
697
     * make mbaff happy, so we can't move all this logic to fill_caches */
698
    if(FRAME_MBAFF){
699
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
700
        const int16_t *mv;
701
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
702
        *C = h->mv_cache[list][scan8[0]-2];
703

    
704
        if(!MB_FIELD
705
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
706
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
707
            if(IS_INTERLACED(mb_types[topright_xy])){
708
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
709
                const int x4 = X4, y4 = Y4;\
710
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
711
                if(!USES_LIST(mb_type,list))\
712
                    return LIST_NOT_USED;\
713
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
714
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
715
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
716
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
717

    
718
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
719
            }
720
        }
721
        if(topright_ref == PART_NOT_AVAILABLE
722
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
723
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
724
            if(!MB_FIELD
725
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
726
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
727
            }
728
            if(MB_FIELD
729
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
730
               && i >= scan8[0]+8){
731
                // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
732
                SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
733
            }
734
        }
735
#undef SET_DIAG_MV
736
    }
737

    
738
    if(topright_ref != PART_NOT_AVAILABLE){
739
        *C= h->mv_cache[list][ i - 8 + part_width ];
740
        return topright_ref;
741
    }else{
742
        tprintf(s->avctx, "topright MV not available\n");
743

    
744
        *C= h->mv_cache[list][ i - 8 - 1 ];
745
        return h->ref_cache[list][ i - 8 - 1 ];
746
    }
747
}
748

    
749
/**
750
 * gets the predicted MV.
751
 * @param n the block index
752
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
753
 * @param mx the x component of the predicted motion vector
754
 * @param my the y component of the predicted motion vector
755
 */
756
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
757
    const int index8= scan8[n];
758
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
759
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
760
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
761
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
762
    const int16_t * C;
763
    int diagonal_ref, match_count;
764

    
765
    assert(part_width==1 || part_width==2 || part_width==4);
766

    
767
/* mv_cache
768
  B . . A T T T T
769
  U . . L . . , .
770
  U . . L . . . .
771
  U . . L . . , .
772
  . . . L . . . .
773
*/
774

    
775
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
776
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
777
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
778
    if(match_count > 1){ //most common
779
        *mx= mid_pred(A[0], B[0], C[0]);
780
        *my= mid_pred(A[1], B[1], C[1]);
781
    }else if(match_count==1){
782
        if(left_ref==ref){
783
            *mx= A[0];
784
            *my= A[1];
785
        }else if(top_ref==ref){
786
            *mx= B[0];
787
            *my= B[1];
788
        }else{
789
            *mx= C[0];
790
            *my= C[1];
791
        }
792
    }else{
793
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
794
            *mx= A[0];
795
            *my= A[1];
796
        }else{
797
            *mx= mid_pred(A[0], B[0], C[0]);
798
            *my= mid_pred(A[1], B[1], C[1]);
799
        }
800
    }
801

    
802
    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);
803
}
804

    
805
/**
806
 * gets the directionally predicted 16x8 MV.
807
 * @param n the block index
808
 * @param mx the x component of the predicted motion vector
809
 * @param my the y component of the predicted motion vector
810
 */
811
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
812
    if(n==0){
813
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
814
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
815

    
816
        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);
817

    
818
        if(top_ref == ref){
819
            *mx= B[0];
820
            *my= B[1];
821
            return;
822
        }
823
    }else{
824
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
825
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
826

    
827
        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);
828

    
829
        if(left_ref == ref){
830
            *mx= A[0];
831
            *my= A[1];
832
            return;
833
        }
834
    }
835

    
836
    //RARE
837
    pred_motion(h, n, 4, list, ref, mx, my);
838
}
839

    
840
/**
841
 * gets the directionally predicted 8x16 MV.
842
 * @param n the block index
843
 * @param mx the x component of the predicted motion vector
844
 * @param my the y component of the predicted motion vector
845
 */
846
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
847
    if(n==0){
848
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
849
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
850

    
851
        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);
852

    
853
        if(left_ref == ref){
854
            *mx= A[0];
855
            *my= A[1];
856
            return;
857
        }
858
    }else{
859
        const int16_t * C;
860
        int diagonal_ref;
861

    
862
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
863

    
864
        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);
865

    
866
        if(diagonal_ref == ref){
867
            *mx= C[0];
868
            *my= C[1];
869
            return;
870
        }
871
    }
872

    
873
    //RARE
874
    pred_motion(h, n, 2, list, ref, mx, my);
875
}
876

    
877
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
878
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
879
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
880

    
881
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
882

    
883
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
884
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
885
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
886

    
887
        *mx = *my = 0;
888
        return;
889
    }
890

    
891
    pred_motion(h, 0, 4, 0, 0, mx, my);
892

    
893
    return;
894
}
895

    
896
static inline void direct_dist_scale_factor(H264Context * const h){
897
    const int poc = h->s.current_picture_ptr->poc;
898
    const int poc1 = h->ref_list[1][0].poc;
899
    int i;
900
    for(i=0; i<h->ref_count[0]; i++){
901
        int poc0 = h->ref_list[0][i].poc;
902
        int td = av_clip(poc1 - poc0, -128, 127);
903
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
904
            h->dist_scale_factor[i] = 256;
905
        }else{
906
            int tb = av_clip(poc - poc0, -128, 127);
907
            int tx = (16384 + (FFABS(td) >> 1)) / td;
908
            h->dist_scale_factor[i] = av_clip((tb*tx + 32) >> 6, -1024, 1023);
909
        }
910
    }
911
    if(FRAME_MBAFF){
912
        for(i=0; i<h->ref_count[0]; i++){
913
            h->dist_scale_factor_field[2*i] =
914
            h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
915
        }
916
    }
917
}
918
static inline void direct_ref_list_init(H264Context * const h){
919
    MpegEncContext * const s = &h->s;
920
    Picture * const ref1 = &h->ref_list[1][0];
921
    Picture * const cur = s->current_picture_ptr;
922
    int list, i, j;
923
    if(cur->pict_type == FF_I_TYPE)
924
        cur->ref_count[0] = 0;
925
    if(cur->pict_type != FF_B_TYPE)
926
        cur->ref_count[1] = 0;
927
    for(list=0; list<2; list++){
928
        cur->ref_count[list] = h->ref_count[list];
929
        for(j=0; j<h->ref_count[list]; j++)
930
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
931
    }
932
    if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
933
        return;
934
    for(list=0; list<2; list++){
935
        for(i=0; i<ref1->ref_count[list]; i++){
936
            const int poc = ref1->ref_poc[list][i];
937
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
938
            for(j=0; j<h->ref_count[list]; j++)
939
                if(h->ref_list[list][j].poc == poc){
940
                    h->map_col_to_list0[list][i] = j;
941
                    break;
942
                }
943
        }
944
    }
945
    if(FRAME_MBAFF){
946
        for(list=0; list<2; list++){
947
            for(i=0; i<ref1->ref_count[list]; i++){
948
                j = h->map_col_to_list0[list][i];
949
                h->map_col_to_list0_field[list][2*i] = 2*j;
950
                h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
951
            }
952
        }
953
    }
954
}
955

    
956
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
957
    MpegEncContext * const s = &h->s;
958
    const int mb_xy =   h->mb_xy;
959
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
960
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
961
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
962
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
963
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
964
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
965
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
966
    const int is_b8x8 = IS_8X8(*mb_type);
967
    unsigned int sub_mb_type;
968
    int i8, i4;
969

    
970
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
971
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
972
        /* FIXME save sub mb types from previous frames (or derive from MVs)
973
         * so we know exactly what block size to use */
974
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
975
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
976
    }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
977
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
978
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
979
    }else{
980
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
981
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
982
    }
983
    if(!is_b8x8)
984
        *mb_type |= MB_TYPE_DIRECT2;
985
    if(MB_FIELD)
986
        *mb_type |= MB_TYPE_INTERLACED;
987

    
988
    tprintf(s->avctx, "mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
989

    
990
    if(h->direct_spatial_mv_pred){
991
        int ref[2];
992
        int mv[2][2];
993
        int list;
994

    
995
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
996

    
997
        /* ref = min(neighbors) */
998
        for(list=0; list<2; list++){
999
            int refa = h->ref_cache[list][scan8[0] - 1];
1000
            int refb = h->ref_cache[list][scan8[0] - 8];
1001
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1002
            if(refc == -2)
1003
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1004
            ref[list] = refa;
1005
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1006
                ref[list] = refb;
1007
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1008
                ref[list] = refc;
1009
            if(ref[list] < 0)
1010
                ref[list] = -1;
1011
        }
1012

    
1013
        if(ref[0] < 0 && ref[1] < 0){
1014
            ref[0] = ref[1] = 0;
1015
            mv[0][0] = mv[0][1] =
1016
            mv[1][0] = mv[1][1] = 0;
1017
        }else{
1018
            for(list=0; list<2; list++){
1019
                if(ref[list] >= 0)
1020
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1021
                else
1022
                    mv[list][0] = mv[list][1] = 0;
1023
            }
1024
        }
1025

    
1026
        if(ref[1] < 0){
1027
            if(!is_b8x8)
1028
                *mb_type &= ~MB_TYPE_L1;
1029
            sub_mb_type &= ~MB_TYPE_L1;
1030
        }else if(ref[0] < 0){
1031
            if(!is_b8x8)
1032
                *mb_type &= ~MB_TYPE_L0;
1033
            sub_mb_type &= ~MB_TYPE_L0;
1034
        }
1035

    
1036
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1037
            int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1038
            int mb_types_col[2];
1039
            int b8_stride = h->b8_stride;
1040
            int b4_stride = h->b_stride;
1041

    
1042
            *mb_type = (*mb_type & ~MB_TYPE_16x16) | MB_TYPE_8x8;
1043

    
1044
            if(IS_INTERLACED(*mb_type)){
1045
                mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1046
                mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1047
                if(s->mb_y&1){
1048
                    l1ref0 -= 2*b8_stride;
1049
                    l1ref1 -= 2*b8_stride;
1050
                    l1mv0 -= 4*b4_stride;
1051
                    l1mv1 -= 4*b4_stride;
1052
                }
1053
                b8_stride *= 3;
1054
                b4_stride *= 6;
1055
            }else{
1056
                int cur_poc = s->current_picture_ptr->poc;
1057
                int *col_poc = h->ref_list[1]->field_poc;
1058
                int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
1059
                int dy = 2*col_parity - (s->mb_y&1);
1060
                mb_types_col[0] =
1061
                mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy + col_parity*s->mb_stride];
1062
                l1ref0 += dy*b8_stride;
1063
                l1ref1 += dy*b8_stride;
1064
                l1mv0 += 2*dy*b4_stride;
1065
                l1mv1 += 2*dy*b4_stride;
1066
                b8_stride = 0;
1067
            }
1068

    
1069
            for(i8=0; i8<4; i8++){
1070
                int x8 = i8&1;
1071
                int y8 = i8>>1;
1072
                int xy8 = x8+y8*b8_stride;
1073
                int xy4 = 3*x8+y8*b4_stride;
1074
                int a=0, b=0;
1075

    
1076
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1077
                    continue;
1078
                h->sub_mb_type[i8] = sub_mb_type;
1079

    
1080
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1081
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1082
                if(!IS_INTRA(mb_types_col[y8])
1083
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1084
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1085
                    if(ref[0] > 0)
1086
                        a= pack16to32(mv[0][0],mv[0][1]);
1087
                    if(ref[1] > 0)
1088
                        b= pack16to32(mv[1][0],mv[1][1]);
1089
                }else{
1090
                    a= pack16to32(mv[0][0],mv[0][1]);
1091
                    b= pack16to32(mv[1][0],mv[1][1]);
1092
                }
1093
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1094
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1095
            }
1096
        }else if(IS_16X16(*mb_type)){
1097
            int a=0, b=0;
1098

    
1099
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1100
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1101
            if(!IS_INTRA(mb_type_col)
1102
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1103
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1104
                       && (h->x264_build>33 || !h->x264_build)))){
1105
                if(ref[0] > 0)
1106
                    a= pack16to32(mv[0][0],mv[0][1]);
1107
                if(ref[1] > 0)
1108
                    b= pack16to32(mv[1][0],mv[1][1]);
1109
            }else{
1110
                a= pack16to32(mv[0][0],mv[0][1]);
1111
                b= pack16to32(mv[1][0],mv[1][1]);
1112
            }
1113
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1114
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1115
        }else{
1116
            for(i8=0; i8<4; i8++){
1117
                const int x8 = i8&1;
1118
                const int y8 = i8>>1;
1119

    
1120
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1121
                    continue;
1122
                h->sub_mb_type[i8] = sub_mb_type;
1123

    
1124
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1125
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1126
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1127
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1128

    
1129
                /* col_zero_flag */
1130
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0
1131
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1132
                                                  && (h->x264_build>33 || !h->x264_build)))){
1133
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1134
                    if(IS_SUB_8X8(sub_mb_type)){
1135
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1136
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1137
                            if(ref[0] == 0)
1138
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1139
                            if(ref[1] == 0)
1140
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1141
                        }
1142
                    }else
1143
                    for(i4=0; i4<4; i4++){
1144
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1145
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1146
                            if(ref[0] == 0)
1147
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1148
                            if(ref[1] == 0)
1149
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1150
                        }
1151
                    }
1152
                }
1153
            }
1154
        }
1155
    }else{ /* direct temporal mv pred */
1156
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1157
        const int *dist_scale_factor = h->dist_scale_factor;
1158

    
1159
        if(FRAME_MBAFF){
1160
            if(IS_INTERLACED(*mb_type)){
1161
                map_col_to_list0[0] = h->map_col_to_list0_field[0];
1162
                map_col_to_list0[1] = h->map_col_to_list0_field[1];
1163
                dist_scale_factor = h->dist_scale_factor_field;
1164
            }
1165
            if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1166
                /* FIXME assumes direct_8x8_inference == 1 */
1167
                const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1168
                int mb_types_col[2];
1169
                int y_shift;
1170

    
1171
                *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1172
                         | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1173
                         | (*mb_type & MB_TYPE_INTERLACED);
1174
                sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1175

    
1176
                if(IS_INTERLACED(*mb_type)){
1177
                    /* frame to field scaling */
1178
                    mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1179
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1180
                    if(s->mb_y&1){
1181
                        l1ref0 -= 2*h->b8_stride;
1182
                        l1ref1 -= 2*h->b8_stride;
1183
                        l1mv0 -= 4*h->b_stride;
1184
                        l1mv1 -= 4*h->b_stride;
1185
                    }
1186
                    y_shift = 0;
1187

    
1188
                    if(   (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1189
                       && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1190
                       && !is_b8x8)
1191
                        *mb_type |= MB_TYPE_16x8;
1192
                    else
1193
                        *mb_type |= MB_TYPE_8x8;
1194
                }else{
1195
                    /* field to frame scaling */
1196
                    /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
1197
                     * but in MBAFF, top and bottom POC are equal */
1198
                    int dy = (s->mb_y&1) ? 1 : 2;
1199
                    mb_types_col[0] =
1200
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1201
                    l1ref0 += dy*h->b8_stride;
1202
                    l1ref1 += dy*h->b8_stride;
1203
                    l1mv0 += 2*dy*h->b_stride;
1204
                    l1mv1 += 2*dy*h->b_stride;
1205
                    y_shift = 2;
1206

    
1207
                    if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1208
                       && !is_b8x8)
1209
                        *mb_type |= MB_TYPE_16x16;
1210
                    else
1211
                        *mb_type |= MB_TYPE_8x8;
1212
                }
1213

    
1214
                for(i8=0; i8<4; i8++){
1215
                    const int x8 = i8&1;
1216
                    const int y8 = i8>>1;
1217
                    int ref0, scale;
1218
                    const int16_t (*l1mv)[2]= l1mv0;
1219

    
1220
                    if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1221
                        continue;
1222
                    h->sub_mb_type[i8] = sub_mb_type;
1223

    
1224
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1225
                    if(IS_INTRA(mb_types_col[y8])){
1226
                        fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1227
                        fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1228
                        fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1229
                        continue;
1230
                    }
1231

    
1232
                    ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1233
                    if(ref0 >= 0)
1234
                        ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1235
                    else{
1236
                        ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1237
                        l1mv= l1mv1;
1238
                    }
1239
                    scale = dist_scale_factor[ref0];
1240
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1241

    
1242
                    {
1243
                        const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
1244
                        int my_col = (mv_col[1]<<y_shift)/2;
1245
                        int mx = (scale * mv_col[0] + 128) >> 8;
1246
                        int my = (scale * my_col + 128) >> 8;
1247
                        fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1248
                        fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1249
                    }
1250
                }
1251
                return;
1252
            }
1253
        }
1254

    
1255
        /* one-to-one mv scaling */
1256

    
1257
        if(IS_16X16(*mb_type)){
1258
            int ref, mv0, mv1;
1259

    
1260
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1261
            if(IS_INTRA(mb_type_col)){
1262
                ref=mv0=mv1=0;
1263
            }else{
1264
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1265
                                                : map_col_to_list0[1][l1ref1[0]];
1266
                const int scale = dist_scale_factor[ref0];
1267
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1268
                int mv_l0[2];
1269
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1270
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1271
                ref= ref0;
1272
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
1273
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1274
            }
1275
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1276
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1277
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1278
        }else{
1279
            for(i8=0; i8<4; i8++){
1280
                const int x8 = i8&1;
1281
                const int y8 = i8>>1;
1282
                int ref0, scale;
1283
                const int16_t (*l1mv)[2]= l1mv0;
1284

    
1285
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1286
                    continue;
1287
                h->sub_mb_type[i8] = sub_mb_type;
1288
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1289
                if(IS_INTRA(mb_type_col)){
1290
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1291
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1292
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1293
                    continue;
1294
                }
1295

    
1296
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1297
                if(ref0 >= 0)
1298
                    ref0 = map_col_to_list0[0][ref0];
1299
                else{
1300
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1301
                    l1mv= l1mv1;
1302
                }
1303
                scale = dist_scale_factor[ref0];
1304

    
1305
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1306
                if(IS_SUB_8X8(sub_mb_type)){
1307
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1308
                    int mx = (scale * mv_col[0] + 128) >> 8;
1309
                    int my = (scale * mv_col[1] + 128) >> 8;
1310
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1311
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1312
                }else
1313
                for(i4=0; i4<4; i4++){
1314
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1315
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1316
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1317
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1318
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1319
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1320
                }
1321
            }
1322
        }
1323
    }
1324
}
1325

    
1326
static inline void write_back_motion(H264Context *h, int mb_type){
1327
    MpegEncContext * const s = &h->s;
1328
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1329
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1330
    int list;
1331

    
1332
    if(!USES_LIST(mb_type, 0))
1333
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1334

    
1335
    for(list=0; list<h->list_count; list++){
1336
        int y;
1337
        if(!USES_LIST(mb_type, list))
1338
            continue;
1339

    
1340
        for(y=0; y<4; y++){
1341
            *(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];
1342
            *(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];
1343
        }
1344
        if( h->pps.cabac ) {
1345
            if(IS_SKIP(mb_type))
1346
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1347
            else
1348
            for(y=0; y<4; y++){
1349
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1350
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1351
            }
1352
        }
1353

    
1354
        {
1355
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1356
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1357
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1358
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1359
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1360
        }
1361
    }
1362

    
1363
    if(h->slice_type == FF_B_TYPE && h->pps.cabac){
1364
        if(IS_8X8(mb_type)){
1365
            uint8_t *direct_table = &h->direct_table[b8_xy];
1366
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1367
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1368
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1369
        }
1370
    }
1371
}
1372

    
1373
/**
1374
 * Decodes a network abstraction layer unit.
1375
 * @param consumed is the number of bytes used as input
1376
 * @param length is the length of the array
1377
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1378
 * @returns decoded bytes, might be src+1 if no escapes
1379
 */
1380
static const uint8_t *decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1381
    int i, si, di;
1382
    uint8_t *dst;
1383
    int bufidx;
1384

    
1385
//    src[0]&0x80;                //forbidden bit
1386
    h->nal_ref_idc= src[0]>>5;
1387
    h->nal_unit_type= src[0]&0x1F;
1388

    
1389
    src++; length--;
1390
#if 0
1391
    for(i=0; i<length; i++)
1392
        printf("%2X ", src[i]);
1393
#endif
1394
    for(i=0; i+1<length; i+=2){
1395
        if(src[i]) continue;
1396
        if(i>0 && src[i-1]==0) i--;
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
    }
1405

    
1406
    if(i>=length-1){ //no escaped 0
1407
        *dst_length= length;
1408
        *consumed= length+1; //+1 for the header
1409
        return src;
1410
    }
1411

    
1412
    bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1413
    h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length);
1414
    dst= h->rbsp_buffer[bufidx];
1415

    
1416
    if (dst == NULL){
1417
        return NULL;
1418
    }
1419

    
1420
//printf("decoding esc\n");
1421
    si=di=0;
1422
    while(si<length){
1423
        //remove escapes (very rare 1:2^22)
1424
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1425
            if(src[si+2]==3){ //escape
1426
                dst[di++]= 0;
1427
                dst[di++]= 0;
1428
                si+=3;
1429
                continue;
1430
            }else //next start code
1431
                break;
1432
        }
1433

    
1434
        dst[di++]= src[si++];
1435
    }
1436

    
1437
    *dst_length= di;
1438
    *consumed= si + 1;//+1 for the header
1439
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1440
    return dst;
1441
}
1442

    
1443
/**
1444
 * identifies the exact end of the bitstream
1445
 * @return the length of the trailing, or 0 if damaged
1446
 */
1447
static int decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1448
    int v= *src;
1449
    int r;
1450

    
1451
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1452

    
1453
    for(r=1; r<9; r++){
1454
        if(v&1) return r;
1455
        v>>=1;
1456
    }
1457
    return 0;
1458
}
1459

    
1460
/**
1461
 * idct tranforms the 16 dc values and dequantize them.
1462
 * @param qp quantization parameter
1463
 */
1464
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1465
#define stride 16
1466
    int i;
1467
    int temp[16]; //FIXME check if this is a good idea
1468
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1469
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1470

    
1471
//memset(block, 64, 2*256);
1472
//return;
1473
    for(i=0; i<4; i++){
1474
        const int offset= y_offset[i];
1475
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1476
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1477
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1478
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1479

    
1480
        temp[4*i+0]= z0+z3;
1481
        temp[4*i+1]= z1+z2;
1482
        temp[4*i+2]= z1-z2;
1483
        temp[4*i+3]= z0-z3;
1484
    }
1485

    
1486
    for(i=0; i<4; i++){
1487
        const int offset= x_offset[i];
1488
        const int z0= temp[4*0+i] + temp[4*2+i];
1489
        const int z1= temp[4*0+i] - temp[4*2+i];
1490
        const int z2= temp[4*1+i] - temp[4*3+i];
1491
        const int z3= temp[4*1+i] + temp[4*3+i];
1492

    
1493
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1494
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1495
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1496
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1497
    }
1498
}
1499

    
1500
#if 0
1501
/**
1502
 * dct tranforms the 16 dc values.
1503
 * @param qp quantization parameter ??? FIXME
1504
 */
1505
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1506
//    const int qmul= dequant_coeff[qp][0];
1507
    int i;
1508
    int temp[16]; //FIXME check if this is a good idea
1509
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1510
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1511

1512
    for(i=0; i<4; i++){
1513
        const int offset= y_offset[i];
1514
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1515
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1516
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1517
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1518

1519
        temp[4*i+0]= z0+z3;
1520
        temp[4*i+1]= z1+z2;
1521
        temp[4*i+2]= z1-z2;
1522
        temp[4*i+3]= z0-z3;
1523
    }
1524

1525
    for(i=0; i<4; i++){
1526
        const int offset= x_offset[i];
1527
        const int z0= temp[4*0+i] + temp[4*2+i];
1528
        const int z1= temp[4*0+i] - temp[4*2+i];
1529
        const int z2= temp[4*1+i] - temp[4*3+i];
1530
        const int z3= temp[4*1+i] + temp[4*3+i];
1531

1532
        block[stride*0 +offset]= (z0 + z3)>>1;
1533
        block[stride*2 +offset]= (z1 + z2)>>1;
1534
        block[stride*8 +offset]= (z1 - z2)>>1;
1535
        block[stride*10+offset]= (z0 - z3)>>1;
1536
    }
1537
}
1538
#endif
1539

    
1540
#undef xStride
1541
#undef stride
1542

    
1543
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1544
    const int stride= 16*2;
1545
    const int xStride= 16;
1546
    int a,b,c,d,e;
1547

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

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

    
1558
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1559
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1560
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1561
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1562
}
1563

    
1564
#if 0
1565
static void chroma_dc_dct_c(DCTELEM *block){
1566
    const int stride= 16*2;
1567
    const int xStride= 16;
1568
    int a,b,c,d,e;
1569

1570
    a= block[stride*0 + xStride*0];
1571
    b= block[stride*0 + xStride*1];
1572
    c= block[stride*1 + xStride*0];
1573
    d= block[stride*1 + xStride*1];
1574

1575
    e= a-b;
1576
    a= a+b;
1577
    b= c-d;
1578
    c= c+d;
1579

1580
    block[stride*0 + xStride*0]= (a+c);
1581
    block[stride*0 + xStride*1]= (e+b);
1582
    block[stride*1 + xStride*0]= (a-c);
1583
    block[stride*1 + xStride*1]= (e-b);
1584
}
1585
#endif
1586

    
1587
/**
1588
 * gets the chroma qp.
1589
 */
1590
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1591
    return h->pps.chroma_qp_table[t][qscale & 0xff];
1592
}
1593

    
1594
//FIXME need to check that this does not overflow signed 32 bit for low qp, I am not sure, it's very close
1595
//FIXME check that gcc inlines this (and optimizes intra & separate_dc stuff away)
1596
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int separate_dc){
1597
    int i;
1598
    const int * const quant_table= quant_coeff[qscale];
1599
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1600
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1601
    const unsigned int threshold2= (threshold1<<1);
1602
    int last_non_zero;
1603

    
1604
    if(separate_dc){
1605
        if(qscale<=18){
1606
            //avoid overflows
1607
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1608
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1609
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1610

    
1611
            int level= block[0]*quant_coeff[qscale+18][0];
1612
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1613
                if(level>0){
1614
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1615
                    block[0]= level;
1616
                }else{
1617
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1618
                    block[0]= -level;
1619
                }
1620
//                last_non_zero = i;
1621
            }else{
1622
                block[0]=0;
1623
            }
1624
        }else{
1625
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1626
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1627
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1628

    
1629
            int level= block[0]*quant_table[0];
1630
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1631
                if(level>0){
1632
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1633
                    block[0]= level;
1634
                }else{
1635
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1636
                    block[0]= -level;
1637
                }
1638
//                last_non_zero = i;
1639
            }else{
1640
                block[0]=0;
1641
            }
1642
        }
1643
        last_non_zero= 0;
1644
        i=1;
1645
    }else{
1646
        last_non_zero= -1;
1647
        i=0;
1648
    }
1649

    
1650
    for(; i<16; i++){
1651
        const int j= scantable[i];
1652
        int level= block[j]*quant_table[j];
1653

    
1654
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1655
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1656
        if(((unsigned)(level+threshold1))>threshold2){
1657
            if(level>0){
1658
                level= (bias + level)>>QUANT_SHIFT;
1659
                block[j]= level;
1660
            }else{
1661
                level= (bias - level)>>QUANT_SHIFT;
1662
                block[j]= -level;
1663
            }
1664
            last_non_zero = i;
1665
        }else{
1666
            block[j]=0;
1667
        }
1668
    }
1669

    
1670
    return last_non_zero;
1671
}
1672

    
1673
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1674
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1675
                           int src_x_offset, int src_y_offset,
1676
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1677
    MpegEncContext * const s = &h->s;
1678
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1679
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1680
    const int luma_xy= (mx&3) + ((my&3)<<2);
1681
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1682
    uint8_t * src_cb, * src_cr;
1683
    int extra_width= h->emu_edge_width;
1684
    int extra_height= h->emu_edge_height;
1685
    int emu=0;
1686
    const int full_mx= mx>>2;
1687
    const int full_my= my>>2;
1688
    const int pic_width  = 16*s->mb_width;
1689
    const int pic_height = 16*s->mb_height >> MB_FIELD;
1690

    
1691
    if(!pic->data[0]) //FIXME this is unacceptable, some senseable error concealment must be done for missing reference frames
1692
        return;
1693

    
1694
    if(mx&7) extra_width -= 3;
1695
    if(my&7) extra_height -= 3;
1696

    
1697
    if(   full_mx < 0-extra_width
1698
       || full_my < 0-extra_height
1699
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
1700
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
1701
        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);
1702
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1703
        emu=1;
1704
    }
1705

    
1706
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1707
    if(!square){
1708
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1709
    }
1710

    
1711
    if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1712

    
1713
    if(MB_FIELD){
1714
        // chroma offset when predicting from a field of opposite parity
1715
        my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1716
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1717
    }
1718
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1719
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1720

    
1721
    if(emu){
1722
        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);
1723
            src_cb= s->edge_emu_buffer;
1724
    }
1725
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1726

    
1727
    if(emu){
1728
        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);
1729
            src_cr= s->edge_emu_buffer;
1730
    }
1731
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1732
}
1733

    
1734
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1735
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1736
                           int x_offset, int y_offset,
1737
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1738
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1739
                           int list0, int list1){
1740
    MpegEncContext * const s = &h->s;
1741
    qpel_mc_func *qpix_op=  qpix_put;
1742
    h264_chroma_mc_func chroma_op= chroma_put;
1743

    
1744
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1745
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1746
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1747
    x_offset += 8*s->mb_x;
1748
    y_offset += 8*(s->mb_y >> MB_FIELD);
1749

    
1750
    if(list0){
1751
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1752
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1753
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1754
                           qpix_op, chroma_op);
1755

    
1756
        qpix_op=  qpix_avg;
1757
        chroma_op= chroma_avg;
1758
    }
1759

    
1760
    if(list1){
1761
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1762
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1763
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1764
                           qpix_op, chroma_op);
1765
    }
1766
}
1767

    
1768
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1769
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1770
                           int x_offset, int y_offset,
1771
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1772
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1773
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1774
                           int list0, int list1){
1775
    MpegEncContext * const s = &h->s;
1776

    
1777
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1778
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1779
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1780
    x_offset += 8*s->mb_x;
1781
    y_offset += 8*(s->mb_y >> MB_FIELD);
1782

    
1783
    if(list0 && list1){
1784
        /* don't optimize for luma-only case, since B-frames usually
1785
         * use implicit weights => chroma too. */
1786
        uint8_t *tmp_cb = s->obmc_scratchpad;
1787
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1788
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1789
        int refn0 = h->ref_cache[0][ scan8[n] ];
1790
        int refn1 = h->ref_cache[1][ scan8[n] ];
1791

    
1792
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1793
                    dest_y, dest_cb, dest_cr,
1794
                    x_offset, y_offset, qpix_put, chroma_put);
1795
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1796
                    tmp_y, tmp_cb, tmp_cr,
1797
                    x_offset, y_offset, qpix_put, chroma_put);
1798

    
1799
        if(h->use_weight == 2){
1800
            int weight0 = h->implicit_weight[refn0][refn1];
1801
            int weight1 = 64 - weight0;
1802
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
1803
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1804
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1805
        }else{
1806
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1807
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1808
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1809
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1810
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1811
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1812
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1813
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1814
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1815
        }
1816
    }else{
1817
        int list = list1 ? 1 : 0;
1818
        int refn = h->ref_cache[list][ scan8[n] ];
1819
        Picture *ref= &h->ref_list[list][refn];
1820
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1821
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
1822
                    qpix_put, chroma_put);
1823

    
1824
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1825
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
1826
        if(h->use_weight_chroma){
1827
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1828
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1829
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1830
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1831
        }
1832
    }
1833
}
1834

    
1835
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1836
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1837
                           int x_offset, int y_offset,
1838
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1839
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1840
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1841
                           int list0, int list1){
1842
    if((h->use_weight==2 && list0 && list1
1843
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1844
       || h->use_weight==1)
1845
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1846
                         x_offset, y_offset, qpix_put, chroma_put,
1847
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1848
    else
1849
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1850
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1851
}
1852

    
1853
static inline void prefetch_motion(H264Context *h, int list){
1854
    /* fetch pixels for estimated mv 4 macroblocks ahead
1855
     * optimized for 64byte cache lines */
1856
    MpegEncContext * const s = &h->s;
1857
    const int refn = h->ref_cache[list][scan8[0]];
1858
    if(refn >= 0){
1859
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1860
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1861
        uint8_t **src= h->ref_list[list][refn].data;
1862
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1863
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
1864
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1865
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1866
    }
1867
}
1868

    
1869
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1870
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1871
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1872
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1873
    MpegEncContext * const s = &h->s;
1874
    const int mb_xy= h->mb_xy;
1875
    const int mb_type= s->current_picture.mb_type[mb_xy];
1876

    
1877
    assert(IS_INTER(mb_type));
1878

    
1879
    prefetch_motion(h, 0);
1880

    
1881
    if(IS_16X16(mb_type)){
1882
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1883
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1884
                &weight_op[0], &weight_avg[0],
1885
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1886
    }else if(IS_16X8(mb_type)){
1887
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1888
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1889
                &weight_op[1], &weight_avg[1],
1890
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1891
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1892
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1893
                &weight_op[1], &weight_avg[1],
1894
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1895
    }else if(IS_8X16(mb_type)){
1896
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1897
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1898
                &weight_op[2], &weight_avg[2],
1899
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1900
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1901
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1902
                &weight_op[2], &weight_avg[2],
1903
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1904
    }else{
1905
        int i;
1906

    
1907
        assert(IS_8X8(mb_type));
1908

    
1909
        for(i=0; i<4; i++){
1910
            const int sub_mb_type= h->sub_mb_type[i];
1911
            const int n= 4*i;
1912
            int x_offset= (i&1)<<2;
1913
            int y_offset= (i&2)<<1;
1914

    
1915
            if(IS_SUB_8X8(sub_mb_type)){
1916
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1917
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1918
                    &weight_op[3], &weight_avg[3],
1919
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1920
            }else if(IS_SUB_8X4(sub_mb_type)){
1921
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1922
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1923
                    &weight_op[4], &weight_avg[4],
1924
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1925
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1926
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1927
                    &weight_op[4], &weight_avg[4],
1928
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1929
            }else if(IS_SUB_4X8(sub_mb_type)){
1930
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1931
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1932
                    &weight_op[5], &weight_avg[5],
1933
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1934
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1935
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1936
                    &weight_op[5], &weight_avg[5],
1937
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1938
            }else{
1939
                int j;
1940
                assert(IS_SUB_4X4(sub_mb_type));
1941
                for(j=0; j<4; j++){
1942
                    int sub_x_offset= x_offset + 2*(j&1);
1943
                    int sub_y_offset= y_offset +   (j&2);
1944
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1945
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1946
                        &weight_op[6], &weight_avg[6],
1947
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1948
                }
1949
            }
1950
        }
1951
    }
1952

    
1953
    prefetch_motion(h, 1);
1954
}
1955

    
1956
static av_cold void decode_init_vlc(void){
1957
    static int done = 0;
1958

    
1959
    if (!done) {
1960
        int i;
1961
        done = 1;
1962

    
1963
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1964
                 &chroma_dc_coeff_token_len [0], 1, 1,
1965
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
1966

    
1967
        for(i=0; i<4; i++){
1968
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1969
                     &coeff_token_len [i][0], 1, 1,
1970
                     &coeff_token_bits[i][0], 1, 1, 1);
1971
        }
1972

    
1973
        for(i=0; i<3; i++){
1974
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1975
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
1976
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
1977
        }
1978
        for(i=0; i<15; i++){
1979
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
1980
                     &total_zeros_len [i][0], 1, 1,
1981
                     &total_zeros_bits[i][0], 1, 1, 1);
1982
        }
1983

    
1984
        for(i=0; i<6; i++){
1985
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
1986
                     &run_len [i][0], 1, 1,
1987
                     &run_bits[i][0], 1, 1, 1);
1988
        }
1989
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1990
                 &run_len [6][0], 1, 1,
1991
                 &run_bits[6][0], 1, 1, 1);
1992
    }
1993
}
1994

    
1995
static void free_tables(H264Context *h){
1996
    int i;
1997
    H264Context *hx;
1998
    av_freep(&h->intra4x4_pred_mode);
1999
    av_freep(&h->chroma_pred_mode_table);
2000
    av_freep(&h->cbp_table);
2001
    av_freep(&h->mvd_table[0]);
2002
    av_freep(&h->mvd_table[1]);
2003
    av_freep(&h->direct_table);
2004
    av_freep(&h->non_zero_count);
2005
    av_freep(&h->slice_table_base);
2006
    h->slice_table= NULL;
2007

    
2008
    av_freep(&h->mb2b_xy);
2009
    av_freep(&h->mb2b8_xy);
2010

    
2011
    for(i = 0; i < MAX_SPS_COUNT; i++)
2012
        av_freep(h->sps_buffers + i);
2013

    
2014
    for(i = 0; i < MAX_PPS_COUNT; i++)
2015
        av_freep(h->pps_buffers + i);
2016

    
2017
    for(i = 0; i < h->s.avctx->thread_count; i++) {
2018
        hx = h->thread_context[i];
2019
        if(!hx) continue;
2020
        av_freep(&hx->top_borders[1]);
2021
        av_freep(&hx->top_borders[0]);
2022
        av_freep(&hx->s.obmc_scratchpad);
2023
    }
2024
}
2025

    
2026
static void init_dequant8_coeff_table(H264Context *h){
2027
    int i,q,x;
2028
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2029
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
2030
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
2031

    
2032
    for(i=0; i<2; i++ ){
2033
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2034
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
2035
            break;
2036
        }
2037

    
2038
        for(q=0; q<52; q++){
2039
            int shift = ff_div6[q];
2040
            int idx = ff_rem6[q];
2041
            for(x=0; x<64; x++)
2042
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2043
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2044
                    h->pps.scaling_matrix8[i][x]) << shift;
2045
        }
2046
    }
2047
}
2048

    
2049
static void init_dequant4_coeff_table(H264Context *h){
2050
    int i,j,q,x;
2051
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2052
    for(i=0; i<6; i++ ){
2053
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
2054
        for(j=0; j<i; j++){
2055
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2056
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2057
                break;
2058
            }
2059
        }
2060
        if(j<i)
2061
            continue;
2062

    
2063
        for(q=0; q<52; q++){
2064
            int shift = ff_div6[q] + 2;
2065
            int idx = ff_rem6[q];
2066
            for(x=0; x<16; x++)
2067
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2068
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2069
                    h->pps.scaling_matrix4[i][x]) << shift;
2070
        }
2071
    }
2072
}
2073

    
2074
static void init_dequant_tables(H264Context *h){
2075
    int i,x;
2076
    init_dequant4_coeff_table(h);
2077
    if(h->pps.transform_8x8_mode)
2078
        init_dequant8_coeff_table(h);
2079
    if(h->sps.transform_bypass){
2080
        for(i=0; i<6; i++)
2081
            for(x=0; x<16; x++)
2082
                h->dequant4_coeff[i][0][x] = 1<<6;
2083
        if(h->pps.transform_8x8_mode)
2084
            for(i=0; i<2; i++)
2085
                for(x=0; x<64; x++)
2086
                    h->dequant8_coeff[i][0][x] = 1<<6;
2087
    }
2088
}
2089

    
2090

    
2091
/**
2092
 * allocates tables.
2093
 * needs width/height
2094
 */
2095
static int alloc_tables(H264Context *h){
2096
    MpegEncContext * const s = &h->s;
2097
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2098
    int x,y;
2099

    
2100
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2101

    
2102
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2103
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
2104
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2105

    
2106
    CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2107
    CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2108
    CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2109
    CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2110

    
2111
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));
2112
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2113

    
2114
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2115
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2116
    for(y=0; y<s->mb_height; y++){
2117
        for(x=0; x<s->mb_width; x++){
2118
            const int mb_xy= x + y*s->mb_stride;
2119
            const int b_xy = 4*x + 4*y*h->b_stride;
2120
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2121

    
2122
            h->mb2b_xy [mb_xy]= b_xy;
2123
            h->mb2b8_xy[mb_xy]= b8_xy;
2124
        }
2125
    }
2126

    
2127
    s->obmc_scratchpad = NULL;
2128

    
2129
    if(!h->dequant4_coeff[0])
2130
        init_dequant_tables(h);
2131

    
2132
    return 0;
2133
fail:
2134
    free_tables(h);
2135
    return -1;
2136
}
2137

    
2138
/**
2139
 * Mimic alloc_tables(), but for every context thread.
2140
 */
2141
static void clone_tables(H264Context *dst, H264Context *src){
2142
    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
2143
    dst->non_zero_count           = src->non_zero_count;
2144
    dst->slice_table              = src->slice_table;
2145
    dst->cbp_table                = src->cbp_table;
2146
    dst->mb2b_xy                  = src->mb2b_xy;
2147
    dst->mb2b8_xy                 = src->mb2b8_xy;
2148
    dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
2149
    dst->mvd_table[0]             = src->mvd_table[0];
2150
    dst->mvd_table[1]             = src->mvd_table[1];
2151
    dst->direct_table             = src->direct_table;
2152

    
2153
    dst->s.obmc_scratchpad = NULL;
2154
    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2155
}
2156

    
2157
/**
2158
 * Init context
2159
 * Allocate buffers which are not shared amongst multiple threads.
2160
 */
2161
static int context_init(H264Context *h){
2162
    CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2163
    CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2164

    
2165
    return 0;
2166
fail:
2167
    return -1; // free_tables will clean up for us
2168
}
2169

    
2170
static av_cold void common_init(H264Context *h){
2171
    MpegEncContext * const s = &h->s;
2172

    
2173
    s->width = s->avctx->width;
2174
    s->height = s->avctx->height;
2175
    s->codec_id= s->avctx->codec->id;
2176

    
2177
    ff_h264_pred_init(&h->hpc, s->codec_id);
2178

    
2179
    h->dequant_coeff_pps= -1;
2180
    s->unrestricted_mv=1;
2181
    s->decode=1; //FIXME
2182

    
2183
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2184
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2185
}
2186

    
2187
static av_cold int decode_init(AVCodecContext *avctx){
2188
    H264Context *h= avctx->priv_data;
2189
    MpegEncContext * const s = &h->s;
2190

    
2191
    MPV_decode_defaults(s);
2192

    
2193
    s->avctx = avctx;
2194
    common_init(h);
2195

    
2196
    s->out_format = FMT_H264;
2197
    s->workaround_bugs= avctx->workaround_bugs;
2198

    
2199
    // set defaults
2200
//    s->decode_mb= ff_h263_decode_mb;
2201
    s->quarter_sample = 1;
2202
    s->low_delay= 1;
2203

    
2204
    if(avctx->codec_id == CODEC_ID_SVQ3)
2205
        avctx->pix_fmt= PIX_FMT_YUVJ420P;
2206
    else
2207
        avctx->pix_fmt= PIX_FMT_YUV420P;
2208

    
2209
    decode_init_vlc();
2210

    
2211
    if(avctx->extradata_size > 0 && avctx->extradata &&
2212
       *(char *)avctx->extradata == 1){
2213
        h->is_avc = 1;
2214
        h->got_avcC = 0;
2215
    } else {
2216
        h->is_avc = 0;
2217
    }
2218

    
2219
    h->thread_context[0] = h;
2220
    return 0;
2221
}
2222

    
2223
static int frame_start(H264Context *h){
2224
    MpegEncContext * const s = &h->s;
2225
    int i;
2226

    
2227
    if(MPV_frame_start(s, s->avctx) < 0)
2228
        return -1;
2229
    ff_er_frame_start(s);
2230
    /*
2231
     * MPV_frame_start uses pict_type to derive key_frame.
2232
     * This is incorrect for H.264; IDR markings must be used.
2233
     * Zero here; IDR markings per slice in frame or fields are OR'd in later.
2234
     * See decode_nal_units().
2235
     */
2236
    s->current_picture_ptr->key_frame= 0;
2237

    
2238
    assert(s->linesize && s->uvlinesize);
2239

    
2240
    for(i=0; i<16; i++){
2241
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2242
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2243
    }
2244
    for(i=0; i<4; i++){
2245
        h->block_offset[16+i]=
2246
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2247
        h->block_offset[24+16+i]=
2248
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2249
    }
2250

    
2251
    /* can't be in alloc_tables because linesize isn't known there.
2252
     * FIXME: redo bipred weight to not require extra buffer? */
2253
    for(i = 0; i < s->avctx->thread_count; i++)
2254
        if(!h->thread_context[i]->s.obmc_scratchpad)
2255
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2256

    
2257
    /* some macroblocks will be accessed before they're available */
2258
    if(FRAME_MBAFF || s->avctx->thread_count > 1)
2259
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
2260

    
2261
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2262
    s->current_picture_ptr->reference= 0;
2263
    return 0;
2264
}
2265

    
2266
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){
2267
    MpegEncContext * const s = &h->s;
2268
    int i;
2269

    
2270
    src_y  -=   linesize;
2271
    src_cb -= uvlinesize;
2272
    src_cr -= uvlinesize;
2273

    
2274
    // There are two lines saved, the line above the the top macroblock of a pair,
2275
    // and the line above the bottom macroblock
2276
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
2277
    for(i=1; i<17; i++){
2278
        h->left_border[i]= src_y[15+i*  linesize];
2279
    }
2280

    
2281
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2282
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2283

    
2284
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2285
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
2286
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
2287
        for(i=1; i<9; i++){
2288
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
2289
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2290
        }
2291
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2292
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2293
    }
2294
}
2295

    
2296
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){
2297
    MpegEncContext * const s = &h->s;
2298
    int temp8, i;
2299
    uint64_t temp64;
2300
    int deblock_left;
2301
    int deblock_top;
2302
    int mb_xy;
2303

    
2304
    if(h->deblocking_filter == 2) {
2305
        mb_xy = h->mb_xy;
2306
        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2307
        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2308
    } else {
2309
        deblock_left = (s->mb_x > 0);
2310
        deblock_top =  (s->mb_y > 0);
2311
    }
2312

    
2313
    src_y  -=   linesize + 1;
2314
    src_cb -= uvlinesize + 1;
2315
    src_cr -= uvlinesize + 1;
2316

    
2317
#define XCHG(a,b,t,xchg)\
2318
t= a;\
2319
if(xchg)\
2320
    a= b;\
2321
b= t;
2322

    
2323
    if(deblock_left){
2324
        for(i = !deblock_top; i<17; i++){
2325
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2326
        }
2327
    }
2328

    
2329
    if(deblock_top){
2330
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2331
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2332
        if(s->mb_x+1 < s->mb_width){
2333
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2334
        }
2335
    }
2336

    
2337
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2338
        if(deblock_left){
2339
            for(i = !deblock_top; i<9; i++){
2340
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
2341
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2342
            }
2343
        }
2344
        if(deblock_top){
2345
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2346
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2347
        }
2348
    }
2349
}
2350

    
2351
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2352
    MpegEncContext * const s = &h->s;
2353
    int i;
2354

    
2355
    src_y  -= 2 *   linesize;
2356
    src_cb -= 2 * uvlinesize;
2357
    src_cr -= 2 * uvlinesize;
2358

    
2359
    // There are two lines saved, the line above the the top macroblock of a pair,
2360
    // and the line above the bottom macroblock
2361
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
2362
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
2363
    for(i=2; i<34; i++){
2364
        h->left_border[i]= src_y[15+i*  linesize];
2365
    }
2366

    
2367
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
2368
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
2369
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
2370
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
2371

    
2372
    if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2373
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
2374
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
2375
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
2376
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
2377
        for(i=2; i<18; i++){
2378
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
2379
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
2380
        }
2381
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
2382
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
2383
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
2384
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
2385
    }
2386
}
2387

    
2388
static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
2389
    MpegEncContext * const s = &h->s;
2390
    int temp8, i;
2391
    uint64_t temp64;
2392
    int deblock_left = (s->mb_x > 0);
2393
    int deblock_top  = (s->mb_y > 1);
2394

    
2395
    tprintf(s->avctx, "xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
2396

    
2397
    src_y  -= 2 *   linesize + 1;
2398
    src_cb -= 2 * uvlinesize + 1;
2399
    src_cr -= 2 * uvlinesize + 1;
2400

    
2401
#define XCHG(a,b,t,xchg)\
2402
t= a;\
2403
if(xchg)\
2404
    a= b;\
2405
b= t;
2406

    
2407
    if(deblock_left){
2408
        for(i = (!deblock_top)<<1; i<34; i++){
2409
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2410
        }
2411
    }
2412

    
2413
    if(deblock_top){
2414
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2415
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2416
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
2417
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
2418
        if(s->mb_x+1 < s->mb_width){
2419
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2420
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
2421
        }
2422
    }
2423

    
2424
    if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2425
        if(deblock_left){
2426
            for(i = (!deblock_top) << 1; i<18; i++){
2427
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
2428
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
2429
            }
2430
        }
2431
        if(deblock_top){
2432
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2433
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2434
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
2435
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
2436
        }
2437
    }
2438
}
2439

    
2440
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2441
    MpegEncContext * const s = &h->s;
2442
    const int mb_x= s->mb_x;
2443
    const int mb_y= s->mb_y;
2444
    const int mb_xy= h->mb_xy;
2445
    const int mb_type= s->current_picture.mb_type[mb_xy];
2446
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2447
    int linesize, uvlinesize /*dct_offset*/;
2448
    int i;
2449
    int *block_offset = &h->block_offset[0];
2450
    const unsigned int bottom = mb_y & 1;
2451
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass), is_h264 = (simple || s->codec_id == CODEC_ID_H264);
2452
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2453
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2454

    
2455
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2456
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2457
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2458

    
2459
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2460
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2461

    
2462
    if (!simple && MB_FIELD) {
2463
        linesize   = h->mb_linesize   = s->linesize * 2;
2464
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2465
        block_offset = &h->block_offset[24];
2466
        if(mb_y&1){ //FIXME move out of this func?
2467
            dest_y -= s->linesize*15;
2468
            dest_cb-= s->uvlinesize*7;
2469
            dest_cr-= s->uvlinesize*7;
2470
        }
2471
        if(FRAME_MBAFF) {
2472
            int list;
2473
            for(list=0; list<h->list_count; list++){
2474
                if(!USES_LIST(mb_type, list))
2475
                    continue;
2476
                if(IS_16X16(mb_type)){
2477
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
2478
                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2479
                }else{
2480
                    for(i=0; i<16; i+=4){
2481
                        //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
2482
                        int ref = h->ref_cache[list][scan8[i]];
2483
                        if(ref >= 0)
2484
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2485
                    }
2486
                }
2487
            }
2488
        }
2489
    } else {
2490
        linesize   = h->mb_linesize   = s->linesize;
2491
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2492
//        dct_offset = s->linesize * 16;
2493
    }
2494

    
2495
    if(transform_bypass){
2496
        idct_dc_add =
2497
        idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2498
    }else if(IS_8x8DCT(mb_type)){
2499
        idct_dc_add = s->dsp.h264_idct8_dc_add;
2500
        idct_add = s->dsp.h264_idct8_add;
2501
    }else{
2502
        idct_dc_add = s->dsp.h264_idct_dc_add;
2503
        idct_add = s->dsp.h264_idct_add;
2504
    }
2505

    
2506
    if(!simple && FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
2507
       && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
2508
        int mbt_y = mb_y&~1;
2509
        uint8_t *top_y  = s->current_picture.data[0] + (mbt_y * 16* s->linesize  ) + mb_x * 16;
2510
        uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
2511
        uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
2512
        xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
2513
    }
2514

    
2515
    if (!simple && IS_INTRA_PCM(mb_type)) {
2516
        unsigned int x, y;
2517

    
2518
        // The pixels are stored in h->mb array in the same order as levels,
2519
        // copy them in output in the correct order.
2520
        for(i=0; i<16; i++) {
2521
            for (y=0; y<4; y++) {
2522
                for (x=0; x<4; x++) {
2523
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
2524
                }
2525
            }
2526
        }
2527
        for(i=16; i<16+4; i++) {
2528
            for (y=0; y<4; y++) {
2529
                for (x=0; x<4; x++) {
2530
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2531
                }
2532
            }
2533
        }
2534
        for(i=20; i<20+4; i++) {
2535
            for (y=0; y<4; y++) {
2536
                for (x=0; x<4; x++) {
2537
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2538
                }
2539
            }
2540
        }
2541
    } else {
2542
        if(IS_INTRA(mb_type)){
2543
            if(h->deblocking_filter && (simple || !FRAME_MBAFF))
2544
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2545

    
2546
            if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2547
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2548
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2549
            }
2550

    
2551
            if(IS_INTRA4x4(mb_type)){
2552
                if(simple || !s->encoding){
2553
                    if(IS_8x8DCT(mb_type)){
2554
                        for(i=0; i<16; i+=4){
2555
                            uint8_t * const ptr= dest_y + block_offset[i];
2556
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2557
                            const int nnz = h->non_zero_count_cache[ scan8[i] ];
2558
                            h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2559
                                                   (h->topright_samples_available<<i)&0x4000, linesize);
2560
                            if(nnz){
2561
                                if(nnz == 1 && h->mb[i*16])
2562
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
2563
                                else
2564
                                    idct_add(ptr, h->mb + i*16, linesize);
2565
                            }
2566
                        }
2567
                    }else
2568
                    for(i=0; i<16; i++){
2569
                        uint8_t * const ptr= dest_y + block_offset[i];
2570
                        uint8_t *topright;
2571
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2572
                        int nnz, tr;
2573

    
2574
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2575
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2576
                            assert(mb_y || linesize <= block_offset[i]);
2577
                            if(!topright_avail){
2578
                                tr= ptr[3 - linesize]*0x01010101;
2579
                                topright= (uint8_t*) &tr;
2580
                            }else
2581
                                topright= ptr + 4 - linesize;
2582
                        }else
2583
                            topright= NULL;
2584

    
2585
                        h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2586
                        nnz = h->non_zero_count_cache[ scan8[i] ];
2587
                        if(nnz){
2588
                            if(is_h264){
2589
                                if(nnz == 1 && h->mb[i*16])
2590
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
2591
                                else
2592
                                    idct_add(ptr, h->mb + i*16, linesize);
2593
                            }else
2594
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2595
                        }
2596
                    }
2597
                }
2598
            }else{
2599
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2600
                if(is_h264){
2601
                    if(!transform_bypass)
2602
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
2603
                }else
2604
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2605
            }
2606
            if(h->deblocking_filter && (simple || !FRAME_MBAFF))
2607
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2608
        }else if(is_h264){
2609
            hl_motion(h, dest_y, dest_cb, dest_cr,
2610
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2611
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2612
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2613
        }
2614

    
2615

    
2616
        if(!IS_INTRA4x4(mb_type)){
2617
            if(is_h264){
2618
                if(IS_INTRA16x16(mb_type)){
2619
                    for(i=0; i<16; i++){
2620
                        if(h->non_zero_count_cache[ scan8[i] ])
2621
                            idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2622
                        else if(h->mb[i*16])
2623
                            idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2624
                    }
2625
                }else{
2626
                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2627
                    for(i=0; i<16; i+=di){
2628
                        int nnz = h->non_zero_count_cache[ scan8[i] ];
2629
                        if(nnz){
2630
                            if(nnz==1 && h->mb[i*16])
2631
                                idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2632
                            else
2633
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2634
                        }
2635
                    }
2636
                }
2637
            }else{
2638
                for(i=0; i<16; i++){
2639
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2640
                        uint8_t * const ptr= dest_y + block_offset[i];
2641
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2642
                    }
2643
                }
2644
            }
2645
        }
2646

    
2647
        if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2648
            uint8_t *dest[2] = {dest_cb, dest_cr};
2649
            if(transform_bypass){
2650
                idct_add = idct_dc_add = s->dsp.add_pixels4;
2651
            }else{
2652
                idct_add = s->dsp.h264_idct_add;
2653
                idct_dc_add = s->dsp.h264_idct_dc_add;
2654
                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]);
2655
                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]);
2656
            }
2657
            if(is_h264){
2658
                for(i=16; i<16+8; i++){
2659
                    if(h->non_zero_count_cache[ scan8[i] ])
2660
                        idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2661
                    else if(h->mb[i*16])
2662
                        idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2663
                }
2664
            }else{
2665
                for(i=16; i<16+8; i++){
2666
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2667
                        uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
2668
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2669
                    }
2670
                }
2671
            }
2672
        }
2673
    }
2674
    if(h->deblocking_filter) {
2675
        if (!simple && FRAME_MBAFF) {
2676
            //FIXME try deblocking one mb at a time?
2677
            // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
2678
            const int mb_y = s->mb_y - 1;
2679
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
2680
            const int mb_xy= mb_x + mb_y*s->mb_stride;
2681
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
2682
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
2683
            if (!bottom) return;
2684
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2685
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2686
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2687

    
2688
            if(IS_INTRA(mb_type_top | mb_type_bottom))
2689
                xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
2690

    
2691
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
2692
            // deblock a pair
2693
            // top
2694
            s->mb_y--; h->mb_xy -= s->mb_stride;
2695
            tprintf(h->s.avctx, "call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
2696
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
2697
            h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2698
            h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2699
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
2700
            // bottom
2701
            s->mb_y++; h->mb_xy += s->mb_stride;
2702
            tprintf(h->s.avctx, "call mbaff filter_mb\n");
2703
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
2704
            h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
2705
            h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
2706
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2707
        } else {
2708
            tprintf(h->s.avctx, "call filter_mb\n");
2709
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
2710
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
2711
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2712
        }
2713
    }
2714
}
2715

    
2716
/**
2717
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2718
 */
2719
static void hl_decode_mb_simple(H264Context *h){
2720
    hl_decode_mb_internal(h, 1);
2721
}
2722

    
2723
/**
2724
 * Process a macroblock; this handles edge cases, such as interlacing.
2725
 */
2726
static void av_noinline hl_decode_mb_complex(H264Context *h){
2727
    hl_decode_mb_internal(h, 0);
2728
}
2729

    
2730
static void hl_decode_mb(H264Context *h){
2731
    MpegEncContext * const s = &h->s;
2732
    const int mb_xy= h->mb_xy;
2733
    const int mb_type= s->current_picture.mb_type[mb_xy];
2734
    int is_complex = FRAME_MBAFF || MB_FIELD || IS_INTRA_PCM(mb_type) || s->codec_id != CODEC_ID_H264 ||
2735
                    (ENABLE_GRAY && (s->flags&CODEC_FLAG_GRAY)) || (ENABLE_H264_ENCODER && s->encoding) || ENABLE_SMALL;
2736

    
2737
    if(ENABLE_H264_ENCODER && !s->decode)
2738
        return;
2739

    
2740
    if (is_complex)
2741
        hl_decode_mb_complex(h);
2742
    else hl_decode_mb_simple(h);
2743
}
2744

    
2745
static void pic_as_field(Picture *pic, const int parity){
2746
    int i;
2747
    for (i = 0; i < 4; ++i) {
2748
        if (parity == PICT_BOTTOM_FIELD)
2749
            pic->data[i] += pic->linesize[i];
2750
        pic->reference = parity;
2751
        pic->linesize[i] *= 2;
2752
    }
2753
}
2754

    
2755
static int split_field_copy(Picture *dest, Picture *src,
2756
                            int parity, int id_add){
2757
    int match = !!(src->reference & parity);
2758

    
2759
    if (match) {
2760
        *dest = *src;
2761
        pic_as_field(dest, parity);
2762
        dest->pic_id *= 2;
2763
        dest->pic_id += id_add;
2764
    }
2765

    
2766
    return match;
2767
}
2768

    
2769
/**
2770
 * Split one reference list into field parts, interleaving by parity
2771
 * as per H.264 spec section 8.2.4.2.5. Output fields have their data pointers
2772
 * set to look at the actual start of data for that field.
2773
 *
2774
 * @param dest output list
2775
 * @param dest_len maximum number of fields to put in dest
2776
 * @param src the source reference list containing fields and/or field pairs
2777
 *            (aka short_ref/long_ref, or
2778
 *             refFrameListXShortTerm/refFrameListLongTerm in spec-speak)
2779
 * @param src_len number of Picture's in source (pairs and unmatched fields)
2780
 * @param parity the parity of the picture being decoded/needing
2781
 *        these ref pics (PICT_{TOP,BOTTOM}_FIELD)
2782
 * @return number of fields placed in dest
2783
 */
2784
static int split_field_half_ref_list(Picture *dest, int dest_len,
2785
                                     Picture *src,  int src_len,  int parity){
2786
    int same_parity   = 1;
2787
    int same_i        = 0;
2788
    int opp_i         = 0;
2789
    int out_i;
2790
    int field_output;
2791

    
2792
    for (out_i = 0; out_i < dest_len; out_i += field_output) {
2793
        if (same_parity && same_i < src_len) {
2794
            field_output = split_field_copy(dest + out_i, src + same_i,
2795
                                            parity, 1);
2796
            same_parity = !field_output;
2797
            same_i++;
2798

    
2799
        } else if (opp_i < src_len) {
2800
            field_output = split_field_copy(dest + out_i, src + opp_i,
2801
                                            PICT_FRAME - parity, 0);
2802
            same_parity = field_output;
2803
            opp_i++;
2804

    
2805
        } else {
2806
            break;
2807
        }
2808
    }
2809

    
2810
    return out_i;
2811
}
2812

    
2813
/**
2814
 * Split the reference frame list into a reference field list.
2815
 * This implements H.264 spec 8.2.4.2.5 for a combined input list.
2816
 * The input list contains both reference field pairs and
2817
 * unmatched reference fields; it is ordered as spec describes
2818
 * RefPicListX for frames in 8.2.4.2.1 and 8.2.4.2.3, except that
2819
 * unmatched field pairs are also present. Conceptually this is equivalent
2820
 * to concatenation of refFrameListXShortTerm with refFrameListLongTerm.
2821
 *
2822
 * @param dest output reference list where ordered fields are to be placed
2823
 * @param dest_len max number of fields to place at dest
2824
 * @param src source reference list, as described above
2825
 * @param src_len number of pictures (pairs and unmatched fields) in src
2826
 * @param parity parity of field being currently decoded
2827
 *        (one of PICT_{TOP,BOTTOM}_FIELD)
2828
 * @param long_i index into src array that holds first long reference picture,
2829
 *        or src_len if no long refs present.
2830
 */
2831
static int split_field_ref_list(Picture *dest, int dest_len,
2832
                                Picture *src,  int src_len,
2833
                                int parity,    int long_i){
2834

    
2835
    int i = split_field_half_ref_list(dest, dest_len, src, long_i, parity);
2836
    dest += i;
2837
    dest_len -= i;
2838

    
2839
    i += split_field_half_ref_list(dest, dest_len, src + long_i,
2840
                                   src_len - long_i, parity);
2841
    return i;
2842
}
2843

    
2844
/**
2845
 * fills the default_ref_list.
2846
 */
2847
static int fill_default_ref_list(H264Context *h){
2848
    MpegEncContext * const s = &h->s;
2849
    int i;
2850
    int smallest_poc_greater_than_current = -1;
2851
    int structure_sel;
2852
    Picture sorted_short_ref[32];
2853
    Picture field_entry_list[2][32];
2854
    Picture *frame_list[2];
2855

    
2856
    if (FIELD_PICTURE) {
2857
        structure_sel = PICT_FRAME;
2858
        frame_list[0] = field_entry_list[0];
2859
        frame_list[1] = field_entry_list[1];
2860
    } else {
2861
        structure_sel = 0;
2862
        frame_list[0] = h->default_ref_list[0];
2863
        frame_list[1] = h->default_ref_list[1];
2864
    }
2865

    
2866
    if(h->slice_type==FF_B_TYPE){
2867
        int list;
2868
        int len[2];
2869
        int short_len[2];
2870
        int out_i;
2871
        int limit= INT_MIN;
2872

    
2873
        /* sort frame according to poc in B slice */
2874
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2875
            int best_i=INT_MIN;
2876
            int best_poc=INT_MAX;
2877

    
2878
            for(i=0; i<h->short_ref_count; i++){
2879
                const int poc= h->short_ref[i]->poc;
2880
                if(poc > limit && poc < best_poc){
2881
                    best_poc= poc;
2882
                    best_i= i;
2883
                }
2884
            }
2885

    
2886
            assert(best_i != INT_MIN);
2887

    
2888
            limit= best_poc;
2889
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2890
            tprintf(h->s.avctx, "sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num);
2891
            if (-1 == smallest_poc_greater_than_current) {
2892
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
2893
                    smallest_poc_greater_than_current = out_i;
2894
                }
2895
            }
2896
        }
2897

    
2898
        tprintf(h->s.avctx, "current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
2899

    
2900
        // find the largest poc
2901
        for(list=0; list<2; list++){
2902
            int index = 0;
2903
            int j= -99;
2904
            int step= list ? -1 : 1;
2905

    
2906
            for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
2907
                int sel;
2908
                while(j<0 || j>= h->short_ref_count){
2909
                    if(j != -99 && step == (list ? -1 : 1))
2910
                        return -1;
2911
                    step = -step;
2912
                    j= smallest_poc_greater_than_current + (step>>1);
2913
                }
2914
                sel = sorted_short_ref[j].reference | structure_sel;
2915
                if(sel != PICT_FRAME) continue;
2916
                frame_list[list][index  ]= sorted_short_ref[j];
2917
                frame_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
2918
            }
2919
            short_len[list] = index;
2920

    
2921
            for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
2922
                int sel;
2923
                if(h->long_ref[i] == NULL) continue;
2924
                sel = h->long_ref[i]->reference | structure_sel;
2925
                if(sel != PICT_FRAME) continue;
2926

    
2927
                frame_list[ list ][index  ]= *h->long_ref[i];
2928
                frame_list[ list ][index++].pic_id= i;
2929
            }
2930
            len[list] = index;
2931
        }
2932

    
2933
        for(list=0; list<2; list++){
2934
            if (FIELD_PICTURE)
2935
                len[list] = split_field_ref_list(h->default_ref_list[list],
2936
                                                 h->ref_count[list],
2937
                                                 frame_list[list],
2938
                                                 len[list],
2939
                                                 s->picture_structure,
2940
                                                 short_len[list]);
2941

    
2942
            // swap the two first elements of L1 when L0 and L1 are identical
2943
            if(list && len[0] > 1 && len[0] == len[1])
2944
                for(i=0; h->default_ref_list[0][i].data[0] == h->default_ref_list[1][i].data[0]; i++)
2945
                    if(i == len[0]){
2946
                        FFSWAP(Picture, h->default_ref_list[1][0], h->default_ref_list[1][1]);
2947
                        break;
2948
                    }
2949

    
2950
            if(len[list] < h->ref_count[ list ])
2951
                memset(&h->default_ref_list[list][len[list]], 0, sizeof(Picture)*(h->ref_count[ list ] - len[list]));
2952
        }
2953

    
2954

    
2955
    }else{
2956
        int index=0;
2957
        int short_len;
2958
        for(i=0; i<h->short_ref_count; i++){
2959
            int sel;
2960
            sel = h->short_ref[i]->reference | structure_sel;
2961
            if(sel != PICT_FRAME) continue;
2962
            frame_list[0][index  ]= *h->short_ref[i];
2963
            frame_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2964
        }
2965
        short_len = index;
2966
        for(i = 0; i < 16; i++){
2967
            int sel;
2968
            if(h->long_ref[i] == NULL) continue;
2969
            sel = h->long_ref[i]->reference | structure_sel;
2970
            if(sel != PICT_FRAME) continue;
2971
            frame_list[0][index  ]= *h->long_ref[i];
2972
            frame_list[0][index++].pic_id= i;
2973
        }
2974

    
2975
        if (FIELD_PICTURE)
2976
            index = split_field_ref_list(h->default_ref_list[0],
2977
                                         h->ref_count[0], frame_list[0],
2978
                                         index, s->picture_structure,
2979
                                         short_len);
2980

    
2981
        if(index < h->ref_count[0])
2982
            memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2983
    }
2984
#ifdef TRACE
2985
    for (i=0; i<h->ref_count[0]; i++) {
2986
        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]);
2987
    }
2988
    if(h->slice_type==FF_B_TYPE){
2989
        for (i=0; i<h->ref_count[1]; i++) {
2990
            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]);
2991
        }
2992
    }
2993
#endif
2994
    return 0;
2995
}
2996

    
2997
static void print_short_term(H264Context *h);
2998
static void print_long_term(H264Context *h);
2999

    
3000
/**
3001
 * Extract structure information about the picture described by pic_num in
3002
 * the current decoding context (frame or field). Note that pic_num is
3003
 * picture number without wrapping (so, 0<=pic_num<max_pic_num).
3004
 * @param pic_num picture number for which to extract structure information
3005
 * @param structure one of PICT_XXX describing structure of picture
3006
 *                      with pic_num
3007
 * @return frame number (short term) or long term index of picture
3008
 *         described by pic_num
3009
 */
3010
static int pic_num_extract(H264Context *h, int pic_num, int *structure){
3011
    MpegEncContext * const s = &h->s;
3012

    
3013
    *structure = s->picture_structure;
3014
    if(FIELD_PICTURE){
3015
        if (!(pic_num & 1))
3016
            /* opposite field */
3017
            *structure ^= PICT_FRAME;
3018
        pic_num >>= 1;
3019
    }
3020

    
3021
    return pic_num;
3022
}
3023

    
3024
static int decode_ref_pic_list_reordering(H264Context *h){
3025
    MpegEncContext * const s = &h->s;
3026
    int list, index, pic_structure;
3027

    
3028
    print_short_term(h);
3029
    print_long_term(h);
3030
    if(h->slice_type==FF_I_TYPE || h->slice_type==FF_SI_TYPE) return 0; //FIXME move before func
3031

    
3032
    for(list=0; list<h->list_count; list++){
3033
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3034

    
3035
        if(get_bits1(&s->gb)){
3036
            int pred= h->curr_pic_num;
3037

    
3038
            for(index=0; ; index++){
3039
                unsigned int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3040
                unsigned int pic_id;
3041
                int i;
3042
                Picture *ref = NULL;
3043

    
3044
                if(reordering_of_pic_nums_idc==3)
3045
                    break;
3046

    
3047
                if(index >= h->ref_count[list]){
3048
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3049
                    return -1;
3050
                }
3051

    
3052
                if(reordering_of_pic_nums_idc<3){
3053
                    if(reordering_of_pic_nums_idc<2){
3054
                        const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3055
                        int frame_num;
3056

    
3057
                        if(abs_diff_pic_num > h->max_pic_num){
3058
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3059
                            return -1;
3060
                        }
3061

    
3062
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3063
                        else                                pred+= abs_diff_pic_num;
3064
                        pred &= h->max_pic_num - 1;
3065

    
3066
                        frame_num = pic_num_extract(h, pred, &pic_structure);
3067

    
3068
                        for(i= h->short_ref_count-1; i>=0; i--){
3069
                            ref = h->short_ref[i];
3070
                            assert(ref->reference);
3071
                            assert(!ref->long_ref);
3072
                            if(ref->data[0] != NULL &&
3073
                                   ref->frame_num == frame_num &&
3074
                                   (ref->reference & pic_structure) &&
3075
                                   ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3076
                                break;
3077
                        }
3078
                        if(i>=0)
3079
                            ref->pic_id= pred;
3080
                    }else{
3081
                        int long_idx;
3082
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3083

    
3084
                        long_idx= pic_num_extract(h, pic_id, &pic_structure);
3085

    
3086
                        if(long_idx>31){
3087
                            av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
3088
                            return -1;
3089
                        }
3090
                        ref = h->long_ref[long_idx];
3091
                        assert(!(ref && !ref->reference));
3092
                        if(ref && (ref->reference & pic_structure)){
3093
                            ref->pic_id= pic_id;
3094
                            assert(ref->long_ref);
3095
                            i=0;
3096
                        }else{
3097
                            i=-1;
3098
                        }
3099
                    }
3100

    
3101
                    if (i < 0) {
3102
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3103
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3104
                    } else {
3105
                        for(i=index; i+1<h->ref_count[list]; i++){
3106
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
3107
                                break;
3108
                        }
3109
                        for(; i > index; i--){
3110
                            h->ref_list[list][i]= h->ref_list[list][i-1];
3111
                        }
3112
                        h->ref_list[list][index]= *ref;
3113
                        if (FIELD_PICTURE){
3114
                            pic_as_field(&h->ref_list[list][index], pic_structure);
3115
                        }
3116
                    }
3117
                }else{
3118
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3119
                    return -1;
3120
                }
3121
            }
3122
        }
3123
    }
3124
    for(list=0; list<h->list_count; list++){
3125
        for(index= 0; index < h->ref_count[list]; index++){
3126
            if(!h->ref_list[list][index].data[0])
3127
                h->ref_list[list][index]= s->current_picture;
3128
        }
3129
    }
3130

    
3131
    if(h->slice_type==FF_B_TYPE && !h->direct_spatial_mv_pred)
3132
        direct_dist_scale_factor(h);
3133
    direct_ref_list_init(h);
3134
    return 0;
3135
}
3136

    
3137
static void fill_mbaff_ref_list(H264Context *h){
3138
    int list, i, j;
3139
    for(list=0; list<2; list++){ //FIXME try list_count
3140
        for(i=0; i<h->ref_count[list]; i++){
3141
            Picture *frame = &h->ref_list[list][i];
3142
            Picture *field = &h->ref_list[list][16+2*i];
3143
            field[0] = *frame;
3144
            for(j=0; j<3; j++)
3145
                field[0].linesize[j] <<= 1;
3146
            field[0].reference = PICT_TOP_FIELD;
3147
            field[1] = field[0];
3148
            for(j=0; j<3; j++)
3149
                field[1].data[j] += frame->linesize[j];
3150
            field[1].reference = PICT_BOTTOM_FIELD;
3151

    
3152
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
3153
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
3154
            for(j=0; j<2; j++){
3155
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
3156
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
3157
            }
3158
        }
3159
    }
3160
    for(j=0; j<h->ref_count[1]; j++){
3161
        for(i=0; i<h->ref_count[0]; i++)
3162
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
3163
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
3164
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
3165
    }
3166
}
3167

    
3168
static int pred_weight_table(H264Context *h){
3169
    MpegEncContext * const s = &h->s;
3170
    int list, i;
3171
    int luma_def, chroma_def;
3172

    
3173
    h->use_weight= 0;
3174
    h->use_weight_chroma= 0;
3175
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3176
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3177
    luma_def = 1<<h->luma_log2_weight_denom;
3178
    chroma_def = 1<<h->chroma_log2_weight_denom;
3179

    
3180
    for(list=0; list<2; list++){
3181
        for(i=0; i<h->ref_count[list]; i++){
3182
            int luma_weight_flag, chroma_weight_flag;
3183

    
3184
            luma_weight_flag= get_bits1(&s->gb);
3185
            if(luma_weight_flag){
3186
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3187
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3188
                if(   h->luma_weight[list][i] != luma_def
3189
                   || h->luma_offset[list][i] != 0)
3190
                    h->use_weight= 1;
3191
            }else{
3192
                h->luma_weight[list][i]= luma_def;
3193
                h->luma_offset[list][i]= 0;
3194
            }
3195

    
3196
            chroma_weight_flag= get_bits1(&s->gb);
3197
            if(chroma_weight_flag){
3198
                int j;
3199
                for(j=0; j<2; j++){
3200
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3201
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3202
                    if(   h->chroma_weight[list][i][j] != chroma_def
3203
                       || h->chroma_offset[list][i][j] != 0)
3204
                        h->use_weight_chroma= 1;
3205
                }
3206
            }else{
3207
                int j;
3208
                for(j=0; j<2; j++){
3209
                    h->chroma_weight[list][i][j]= chroma_def;
3210
                    h->chroma_offset[list][i][j]= 0;
3211
                }
3212
            }
3213
        }
3214
        if(h->slice_type != FF_B_TYPE) break;
3215
    }
3216
    h->use_weight= h->use_weight || h->use_weight_chroma;
3217
    return 0;
3218
}
3219

    
3220
static void implicit_weight_table(H264Context *h){
3221
    MpegEncContext * const s = &h->s;
3222
    int ref0, ref1;
3223
    int cur_poc = s->current_picture_ptr->poc;
3224

    
3225
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3226
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3227
        h->use_weight= 0;
3228
        h->use_weight_chroma= 0;
3229
        return;
3230
    }
3231

    
3232
    h->use_weight= 2;
3233
    h->use_weight_chroma= 2;
3234
    h->luma_log2_weight_denom= 5;
3235
    h->chroma_log2_weight_denom= 5;
3236

    
3237
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3238
        int poc0 = h->ref_list[0][ref0].poc;
3239
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3240
            int poc1 = h->ref_list[1][ref1].poc;
3241
            int td = av_clip(poc1 - poc0, -128, 127);
3242
            if(td){
3243
                int tb = av_clip(cur_poc - poc0, -128, 127);
3244
                int tx = (16384 + (FFABS(td) >> 1)) / td;
3245
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3246
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3247
                    h->implicit_weight[ref0][ref1] = 32;
3248
                else
3249
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3250
            }else
3251
                h->implicit_weight[ref0][ref1] = 32;
3252
        }
3253
    }
3254
}
3255

    
3256
/**
3257
 * Mark a picture as no longer needed for reference. The refmask
3258
 * argument allows unreferencing of individual fields or the whole frame.
3259
 * If the picture becomes entirely unreferenced, but is being held for
3260
 * display purposes, it is marked as such.
3261
 * @param refmask mask of fields to unreference; the mask is bitwise
3262
 *                anded with the reference marking of pic
3263
 * @return non-zero if pic becomes entirely unreferenced (except possibly
3264
 *         for display purposes) zero if one of the fields remains in
3265
 *         reference
3266
 */
3267
static inline int unreference_pic(H264Context *h, Picture *pic, int refmask){
3268
    int i;
3269
    if (pic->reference &= refmask) {
3270
        return 0;
3271
    } else {
3272
        for(i = 0; h->delayed_pic[i]; i++)
3273
            if(pic == h->delayed_pic[i]){
3274
                pic->reference=DELAYED_PIC_REF;
3275
                break;
3276
            }
3277
        return 1;
3278
    }
3279
}
3280

    
3281
/**
3282
 * instantaneous decoder refresh.
3283
 */
3284
static void idr(H264Context *h){
3285
    int i;
3286

    
3287
    for(i=0; i<16; i++){
3288
        if (h->long_ref[i] != NULL) {
3289
            unreference_pic(h, h->long_ref[i], 0);
3290
            h->long_ref[i]= NULL;
3291
        }
3292
    }
3293
    h->long_ref_count=0;
3294

    
3295
    for(i=0; i<h->short_ref_count; i++){
3296
        unreference_pic(h, h->short_ref[i], 0);
3297
        h->short_ref[i]= NULL;
3298
    }
3299
    h->short_ref_count=0;
3300
}
3301

    
3302
/* forget old pics after a seek */
3303
static void flush_dpb(AVCodecContext *avctx){
3304
    H264Context *h= avctx->priv_data;
3305
    int i;
3306
    for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
3307
        if(h->delayed_pic[i])
3308
            h->delayed_pic[i]->reference= 0;
3309
        h->delayed_pic[i]= NULL;
3310
    }
3311
    h->outputed_poc= INT_MIN;
3312
    idr(h);
3313
    if(h->s.current_picture_ptr)
3314
        h->s.current_picture_ptr->reference= 0;
3315
    h->s.first_field= 0;
3316
    ff_mpeg_flush(avctx);
3317
}
3318

    
3319
/**
3320
 * Find a Picture in the short term reference list by frame number.
3321
 * @param frame_num frame number to search for
3322
 * @param idx the index into h->short_ref where returned picture is found
3323
 *            undefined if no picture found.
3324
 * @return pointer to the found picture, or NULL if no pic with the provided
3325
 *                 frame number is found
3326
 */
3327
static Picture * find_short(H264Context *h, int frame_num, int *idx){
3328
    MpegEncContext * const s = &h->s;
3329
    int i;
3330

    
3331
    for(i=0; i<h->short_ref_count; i++){
3332
        Picture *pic= h->short_ref[i];
3333
        if(s->avctx->debug&FF_DEBUG_MMCO)
3334
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3335
        if(pic->frame_num == frame_num) {
3336
            *idx = i;
3337
            return pic;
3338
        }
3339
    }
3340
    return NULL;
3341
}
3342

    
3343
/**
3344
 * Remove a picture from the short term reference list by its index in
3345
 * that list.  This does no checking on the provided index; it is assumed
3346
 * to be valid. Other list entries are shifted down.
3347
 * @param i index into h->short_ref of picture to remove.
3348
 */
3349
static void remove_short_at_index(H264Context *h, int i){
3350
    assert(i >= 0 && i < h->short_ref_count);
3351
    h->short_ref[i]= NULL;
3352
    if (--h->short_ref_count)
3353
        memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i)*sizeof(Picture*));
3354
}
3355

    
3356
/**
3357
 *
3358
 * @return the removed picture or NULL if an error occurs
3359
 */
3360
static Picture * remove_short(H264Context *h, int frame_num){
3361
    MpegEncContext * const s = &h->s;
3362
    Picture *pic;
3363
    int i;
3364

    
3365
    if(s->avctx->debug&FF_DEBUG_MMCO)
3366
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3367

    
3368
    pic = find_short(h, frame_num, &i);
3369
    if (pic)
3370
        remove_short_at_index(h, i);
3371

    
3372
    return pic;
3373
}
3374

    
3375
/**
3376
 * Remove a picture from the long term reference list by its index in
3377
 * that list.  This does no checking on the provided index; it is assumed
3378
 * to be valid. The removed entry is set to NULL. Other entries are unaffected.
3379
 * @param i index into h->long_ref of picture to remove.
3380
 */
3381
static void remove_long_at_index(H264Context *h, int i){
3382
    h->long_ref[i]= NULL;
3383
    h->long_ref_count--;
3384
}
3385

    
3386
/**
3387
 *
3388
 * @return the removed picture or NULL if an error occurs
3389
 */
3390
static Picture * remove_long(H264Context *h, int i){
3391
    Picture *pic;
3392

    
3393
    pic= h->long_ref[i];
3394
    if (pic)
3395
        remove_long_at_index(h, i);
3396

    
3397
    return pic;
3398
}
3399

    
3400
/**
3401
 * print short term list
3402
 */
3403
static void print_short_term(H264Context *h) {
3404
    uint32_t i;
3405
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3406
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3407
        for(i=0; i<h->short_ref_count; i++){
3408
            Picture *pic= h->short_ref[i];
3409
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3410
        }
3411
    }
3412
}
3413

    
3414
/**
3415
 * print long term list
3416
 */
3417
static void print_long_term(H264Context *h) {
3418
    uint32_t i;
3419
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3420
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3421
        for(i = 0; i < 16; i++){
3422
            Picture *pic= h->long_ref[i];
3423
            if (pic) {
3424
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3425
            }
3426
        }
3427
    }
3428
}
3429

    
3430
/**
3431
 * Executes the reference picture marking (memory management control operations).
3432
 */
3433
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3434
    MpegEncContext * const s = &h->s;
3435
    int i, j;
3436
    int current_ref_assigned=0;
3437
    Picture *pic;
3438

    
3439
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3440
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3441

    
3442
    for(i=0; i<mmco_count; i++){
3443
        int structure, frame_num, unref_pic;
3444
        if(s->avctx->debug&FF_DEBUG_MMCO)
3445
            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);
3446

    
3447
        switch(mmco[i].opcode){
3448
        case MMCO_SHORT2UNUSED:
3449
            if(s->avctx->debug&FF_DEBUG_MMCO)
3450
                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);
3451
            frame_num = pic_num_extract(h, mmco[i].short_pic_num, &structure);
3452
            pic = find_short(h, frame_num, &j);
3453
            if (pic) {
3454
                if (unreference_pic(h, pic, structure ^ PICT_FRAME))
3455
                    remove_short_at_index(h, j);
3456
            } else if(s->avctx->debug&FF_DEBUG_MMCO)
3457
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref short failure\n");
3458
            break;
3459
        case MMCO_SHORT2LONG:
3460
            if (FIELD_PICTURE && mmco[i].long_arg < h->long_ref_count &&
3461
                    h->long_ref[mmco[i].long_arg]->frame_num ==
3462
                                              mmco[i].short_pic_num / 2) {
3463
                /* do nothing, we've already moved this field pair. */
3464
            } else {
3465
                int frame_num = mmco[i].short_pic_num >> FIELD_PICTURE;
3466

    
3467
                pic= remove_long(h, mmco[i].long_arg);
3468
                if(pic) unreference_pic(h, pic, 0);
3469

    
3470
                h->long_ref[ mmco[i].long_arg ]= remove_short(h, frame_num);
3471
                if (h->long_ref[ mmco[i].long_arg ]){
3472
                    h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3473
                    h->long_ref_count++;
3474
                }
3475
            }
3476
            break;
3477
        case MMCO_LONG2UNUSED:
3478
            j = pic_num_extract(h, mmco[i].long_arg, &structure);
3479
            pic = h->long_ref[j];
3480
            if (pic) {
3481
                if (unreference_pic(h, pic, structure ^ PICT_FRAME))
3482
                    remove_long_at_index(h, j);
3483
            } else if(s->avctx->debug&FF_DEBUG_MMCO)
3484
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref long failure\n");
3485
            break;
3486
        case MMCO_LONG:
3487
            unref_pic = 1;
3488
            if (FIELD_PICTURE && !s->first_field) {
3489
                if (h->long_ref[mmco[i].long_arg] == s->current_picture_ptr) {
3490
                    /* Just mark second field as referenced */
3491
                    unref_pic = 0;
3492
                } else if (s->current_picture_ptr->reference) {
3493
                    /* First field in pair is in short term list or
3494
                     * at a different long term index.
3495
                     * This is not allowed; see 7.4.3, notes 2 and 3.
3496
                     * Report the problem and keep the pair where it is,
3497
                     * and mark this field valid.
3498
                     */
3499
                    av_log(h->s.avctx, AV_LOG_ERROR,
3500
                        "illegal long term reference assignment for second "
3501
                        "field in complementary field pair (first field is "
3502
                        "short term or has non-matching long index)\n");
3503
                    unref_pic = 0;
3504
                }
3505
            }
3506

    
3507
            if (unref_pic) {
3508
                pic= remove_long(h, mmco[i].long_arg);
3509
                if(pic) unreference_pic(h, pic, 0);
3510

    
3511
                h->long_ref[ mmco[i].long_arg ]= s->current_picture_ptr;
3512
                h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3513
                h->long_ref_count++;
3514
            }
3515

    
3516
            s->current_picture_ptr->reference |= s->picture_structure;
3517
            current_ref_assigned=1;
3518
            break;
3519
        case MMCO_SET_MAX_LONG:
3520
            assert(mmco[i].long_arg <= 16);
3521
            // just remove the long term which index is greater than new max
3522
            for(j = mmco[i].long_arg; j<16; j++){
3523
                pic = remove_long(h, j);
3524
                if (pic) unreference_pic(h, pic, 0);
3525
            }
3526
            break;
3527
        case MMCO_RESET:
3528
            while(h->short_ref_count){
3529
                pic= remove_short(h, h->short_ref[0]->frame_num);
3530
                if(pic) unreference_pic(h, pic, 0);
3531
            }
3532
            for(j = 0; j < 16; j++) {
3533
                pic= remove_long(h, j);
3534
                if(pic) unreference_pic(h, pic, 0);
3535
            }
3536
            break;
3537
        default: assert(0);
3538
        }
3539
    }
3540

    
3541
    if (!current_ref_assigned && FIELD_PICTURE &&
3542
            !s->first_field && s->current_picture_ptr->reference) {
3543

    
3544
        /* Second field of complementary field pair; the first field of
3545
         * which is already referenced. If short referenced, it
3546
         * should be first entry in short_ref. If not, it must exist
3547
         * in long_ref; trying to put it on the short list here is an
3548
         * error in the encoded bit stream (ref: 7.4.3, NOTE 2 and 3).
3549
         */
3550
        if (h->short_ref_count && h->short_ref[0] == s->current_picture_ptr) {
3551
            /* Just mark the second field valid */
3552
            s->current_picture_ptr->reference = PICT_FRAME;
3553
        } else if (s->current_picture_ptr->long_ref) {
3554
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term reference "
3555
                                             "assignment for second field "
3556
                                             "in complementary field pair "
3557
                                             "(first field is long term)\n");
3558
        } else {
3559
            /*
3560
             * First field in reference, but not in any sensible place on our
3561
             * reference lists. This shouldn't happen unless reference
3562
             * handling somewhere else is wrong.
3563
             */
3564
            assert(0);
3565
        }
3566
        current_ref_assigned = 1;
3567
    }
3568

    
3569
    if(!current_ref_assigned){
3570
        pic= remove_short(h, s->current_picture_ptr->frame_num);
3571
        if(pic){
3572
            unreference_pic(h, pic, 0);
3573
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3574
        }
3575

    
3576
        if(h->short_ref_count)
3577
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3578

    
3579
        h->short_ref[0]= s->current_picture_ptr;
3580
        h->short_ref[0]->long_ref=0;
3581
        h->short_ref_count++;
3582
        s->current_picture_ptr->reference |= s->picture_structure;
3583
    }
3584

    
3585
    if (h->long_ref_count + h->short_ref_count > h->sps.ref_frame_count){
3586

    
3587
        /* We have too many reference frames, probably due to corrupted
3588
         * stream. Need to discard one frame. Prevents overrun of the
3589
         * short_ref and long_ref buffers.
3590
         */
3591
        av_log(h->s.avctx, AV_LOG_ERROR,
3592
               "number of reference frames exceeds max (probably "
3593
               "corrupt input), discarding one\n");
3594

    
3595
        if (h->long_ref_count) {
3596
            for (i = 0; i < 16; ++i)
3597
                if (h->long_ref[i])
3598
                    break;
3599

    
3600
            assert(i < 16);
3601
            pic = h->long_ref[i];
3602
            remove_long_at_index(h, i);
3603
        } else {
3604
            pic = h->short_ref[h->short_ref_count - 1];
3605
            remove_short_at_index(h, h->short_ref_count - 1);
3606
        }
3607
        unreference_pic(h, pic, 0);
3608
    }
3609

    
3610
    print_short_term(h);
3611
    print_long_term(h);
3612
    return 0;
3613
}
3614

    
3615
static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
3616
    MpegEncContext * const s = &h->s;
3617
    int i;
3618

    
3619
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3620
        s->broken_link= get_bits1(gb) -1;
3621
        h->mmco[0].long_arg= get_bits1(gb) - 1; // current_long_term_idx
3622
        if(h->mmco[0].long_arg == -1)
3623
            h->mmco_index= 0;
3624
        else{
3625
            h->mmco[0].opcode= MMCO_LONG;
3626
            h->mmco_index= 1;
3627
        }
3628
    }else{
3629
        if(get_bits1(gb)){ // adaptive_ref_pic_marking_mode_flag
3630
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
3631
                MMCOOpcode opcode= get_ue_golomb(gb);
3632

    
3633
                h->mmco[i].opcode= opcode;
3634
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3635
                    h->mmco[i].short_pic_num= (h->curr_pic_num - get_ue_golomb(gb) - 1) & (h->max_pic_num - 1);
3636
/*                    if(h->mmco[i].short_pic_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_pic_num ] == NULL){
3637
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
3638
                        return -1;
3639
                    }*/
3640
                }
3641
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3642
                    unsigned int long_arg= get_ue_golomb(gb);
3643
                    if(long_arg >= 32 || (long_arg >= 16 && !(opcode == MMCO_LONG2UNUSED && FIELD_PICTURE))){
3644
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3645
                        return -1;
3646
                    }
3647
                    h->mmco[i].long_arg= long_arg;
3648
                }
3649

    
3650
                if(opcode > (unsigned)MMCO_LONG){
3651
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3652
                    return -1;
3653
                }
3654
                if(opcode == MMCO_END)
3655
                    break;
3656
            }
3657
            h->mmco_index= i;
3658
        }else{
3659
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3660

    
3661
            if(h->short_ref_count && h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count &&
3662
                    !(FIELD_PICTURE && !s->first_field && s->current_picture_ptr->reference)) {
3663
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3664
                h->mmco[0].short_pic_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3665
                h->mmco_index= 1;
3666
                if (FIELD_PICTURE) {
3667
                    h->mmco[0].short_pic_num *= 2;
3668
                    h->mmco[1].opcode= MMCO_SHORT2UNUSED;
3669
                    h->mmco[1].short_pic_num= h->mmco[0].short_pic_num + 1;
3670
                    h->mmco_index= 2;
3671
                }
3672
            }else
3673
                h->mmco_index= 0;
3674
        }
3675
    }
3676

    
3677
    return 0;
3678
}
3679

    
3680
static int init_poc(H264Context *h){
3681
    MpegEncContext * const s = &h->s;
3682
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3683
    int field_poc[2];
3684

    
3685
    if(h->nal_unit_type == NAL_IDR_SLICE){
3686
        h->frame_num_offset= 0;
3687
    }else{
3688
        if(h->frame_num < h->prev_frame_num)
3689
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3690
        else
3691
            h->frame_num_offset= h->prev_frame_num_offset;
3692
    }
3693

    
3694
    if(h->sps.poc_type==0){
3695
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3696

    
3697
        if(h->nal_unit_type == NAL_IDR_SLICE){
3698
             h->prev_poc_msb=
3699
             h->prev_poc_lsb= 0;
3700
        }
3701

    
3702
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3703
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3704
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3705
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3706
        else
3707
            h->poc_msb = h->prev_poc_msb;
3708
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3709
        field_poc[0] =
3710
        field_poc[1] = h->poc_msb + h->poc_lsb;
3711
        if(s->picture_structure == PICT_FRAME)
3712
            field_poc[1] += h->delta_poc_bottom;
3713
    }else if(h->sps.poc_type==1){
3714
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3715
        int i;
3716

    
3717
        if(h->sps.poc_cycle_length != 0)
3718
            abs_frame_num = h->frame_num_offset + h->frame_num;
3719
        else
3720
            abs_frame_num = 0;
3721

    
3722
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
3723
            abs_frame_num--;
3724

    
3725
        expected_delta_per_poc_cycle = 0;
3726
        for(i=0; i < h->sps.poc_cycle_length; i++)
3727
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3728

    
3729
        if(abs_frame_num > 0){
3730
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3731
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3732

    
3733
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3734
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
3735
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3736
        } else
3737
            expectedpoc = 0;
3738

    
3739
        if(h->nal_ref_idc == 0)
3740
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3741

    
3742
        field_poc[0] = expectedpoc + h->delta_poc[0];
3743
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3744

    
3745
        if(s->picture_structure == PICT_FRAME)
3746
            field_poc[1] += h->delta_poc[1];
3747
    }else{
3748
        int poc;
3749
        if(h->nal_unit_type == NAL_IDR_SLICE){
3750
            poc= 0;
3751
        }else{
3752
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3753
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3754
        }
3755
        field_poc[0]= poc;
3756
        field_poc[1]= poc;
3757
    }
3758

    
3759
    if(s->picture_structure != PICT_BOTTOM_FIELD) {
3760
        s->current_picture_ptr->field_poc[0]= field_poc[0];
3761
        s->current_picture_ptr->poc = field_poc[0];
3762
    }
3763
    if(s->picture_structure != PICT_TOP_FIELD) {
3764
        s->current_picture_ptr->field_poc[1]= field_poc[1];
3765
        s->current_picture_ptr->poc = field_poc[1];
3766
    }
3767
    if(!FIELD_PICTURE || !s->first_field) {
3768
        Picture *cur = s->current_picture_ptr;
3769
        cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
3770
    }
3771

    
3772
    return 0;
3773
}
3774

    
3775

    
3776
/**
3777
 * initialize scan tables
3778
 */
3779
static void init_scan_tables(H264Context *h){
3780
    MpegEncContext * const s = &h->s;
3781
    int i;
3782
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
3783
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
3784
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
3785
    }else{
3786
        for(i=0; i<16; i++){
3787
#define T(x) (x>>2) | ((x<<2) & 0xF)
3788
            h->zigzag_scan[i] = T(zigzag_scan[i]);
3789
            h-> field_scan[i] = T( field_scan[i]);
3790
#undef T
3791
        }
3792
    }
3793
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
3794
        memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));
3795
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
3796
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
3797
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
3798
    }else{
3799
        for(i=0; i<64; i++){
3800
#define T(x) (x>>3) | ((x&7)<<3)
3801
            h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);
3802
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
3803
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
3804
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
3805
#undef T
3806
        }
3807
    }
3808
    if(h->sps.transform_bypass){ //FIXME same ugly
3809
        h->zigzag_scan_q0          = zigzag_scan;
3810
        h->zigzag_scan8x8_q0       = zigzag_scan8x8;
3811
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
3812
        h->field_scan_q0           = field_scan;
3813
        h->field_scan8x8_q0        = field_scan8x8;
3814
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
3815
    }else{
3816
        h->zigzag_scan_q0          = h->zigzag_scan;
3817
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
3818
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
3819
        h->field_scan_q0           = h->field_scan;
3820
        h->field_scan8x8_q0        = h->field_scan8x8;
3821
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
3822
    }
3823
}
3824

    
3825
/**
3826
 * Replicates H264 "master" context to thread contexts.
3827
 */
3828
static void clone_slice(H264Context *dst, H264Context *src)
3829
{
3830
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
3831
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
3832
    dst->s.current_picture      = src->s.current_picture;
3833
    dst->s.linesize             = src->s.linesize;
3834
    dst->s.uvlinesize           = src->s.uvlinesize;
3835
    dst->s.first_field          = src->s.first_field;
3836

    
3837
    dst->prev_poc_msb           = src->prev_poc_msb;
3838
    dst->prev_poc_lsb           = src->prev_poc_lsb;
3839
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
3840
    dst->prev_frame_num         = src->prev_frame_num;
3841
    dst->short_ref_count        = src->short_ref_count;
3842

    
3843
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
3844
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
3845
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
3846
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
3847

    
3848
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
3849
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
3850
}
3851

    
3852
/**
3853
 * decodes a slice header.
3854
 * This will also call MPV_common_init() and frame_start() as needed.
3855
 *
3856
 * @param h h264context
3857
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
3858
 *
3859
 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3860
 */
3861
static int decode_slice_header(H264Context *h, H264Context *h0){
3862
    MpegEncContext * const s = &h->s;
3863
    MpegEncContext * const s0 = &h0->s;
3864
    unsigned int first_mb_in_slice;
3865
    unsigned int pps_id;
3866
    int num_ref_idx_active_override_flag;
3867
    static const uint8_t slice_type_map[5]= {FF_P_TYPE, FF_B_TYPE, FF_I_TYPE, FF_SP_TYPE, FF_SI_TYPE};
3868
    unsigned int slice_type, tmp, i;
3869
    int default_ref_list_done = 0;
3870
    int last_pic_structure;
3871

    
3872
    s->dropable= h->nal_ref_idc == 0;
3873

    
3874
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
3875
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
3876
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
3877
    }else{
3878
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
3879
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
3880
    }
3881

    
3882
    first_mb_in_slice= get_ue_golomb(&s->gb);
3883

    
3884
    if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
3885
        h0->current_slice = 0;
3886
        if (!s0->first_field)
3887
            s->current_picture_ptr= NULL;
3888
    }
3889

    
3890
    slice_type= get_ue_golomb(&s->gb);
3891
    if(slice_type > 9){
3892
        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);
3893
        return -1;
3894
    }
3895
    if(slice_type > 4){
3896
        slice_type -= 5;
3897
        h->slice_type_fixed=1;
3898
    }else
3899
        h->slice_type_fixed=0;
3900

    
3901
    slice_type= slice_type_map[ slice_type ];
3902
    if (slice_type == FF_I_TYPE
3903
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
3904
        default_ref_list_done = 1;
3905
    }
3906
    h->slice_type= slice_type;
3907

    
3908
    s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
3909
    if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
3910
        av_log(h->s.avctx, AV_LOG_ERROR,
3911
               "B picture before any references, skipping\n");
3912
        return -1;
3913
    }
3914

    
3915
    pps_id= get_ue_golomb(&s->gb);
3916
    if(pps_id>=MAX_PPS_COUNT){
3917
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3918
        return -1;
3919
    }
3920
    if(!h0->pps_buffers[pps_id]) {
3921
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3922
        return -1;
3923
    }
3924
    h->pps= *h0->pps_buffers[pps_id];
3925

    
3926
    if(!h0->sps_buffers[h->pps.sps_id]) {
3927
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3928
        return -1;
3929
    }
3930
    h->sps = *h0->sps_buffers[h->pps.sps_id];
3931

    
3932
    if(h == h0 && h->dequant_coeff_pps != pps_id){
3933
        h->dequant_coeff_pps = pps_id;
3934
        init_dequant_tables(h);
3935
    }
3936

    
3937
    s->mb_width= h->sps.mb_width;
3938
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3939

    
3940
    h->b_stride=  s->mb_width*4;
3941
    h->b8_stride= s->mb_width*2;
3942

    
3943
    s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
3944
    if(h->sps.frame_mbs_only_flag)
3945
        s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
3946
    else
3947
        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
3948

    
3949
    if (s->context_initialized
3950
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3951
        if(h != h0)
3952
            return -1;   // width / height changed during parallelized decoding
3953
        free_tables(h);
3954
        MPV_common_end(s);
3955
    }
3956
    if (!s->context_initialized) {
3957
        if(h != h0)
3958
            return -1;  // we cant (re-)initialize context during parallel decoding
3959
        if (MPV_common_init(s) < 0)
3960
            return -1;
3961
        s->first_field = 0;
3962

    
3963
        init_scan_tables(h);
3964
        alloc_tables(h);
3965

    
3966
        for(i = 1; i < s->avctx->thread_count; i++) {
3967
            H264Context *c;
3968
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
3969
            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
3970
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
3971
            c->sps = h->sps;
3972
            c->pps = h->pps;
3973
            init_scan_tables(c);
3974
            clone_tables(c, h);
3975
        }
3976

    
3977
        for(i = 0; i < s->avctx->thread_count; i++)
3978
            if(context_init(h->thread_context[i]) < 0)
3979
                return -1;
3980

    
3981
        s->avctx->width = s->width;
3982
        s->avctx->height = s->height;
3983
        s->avctx->sample_aspect_ratio= h->sps.sar;
3984
        if(!s->avctx->sample_aspect_ratio.den)
3985
            s->avctx->sample_aspect_ratio.den = 1;
3986

    
3987
        if(h->sps.timing_info_present_flag){
3988
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale};
3989
            if(h->x264_build > 0 && h->x264_build < 44)
3990
                s->avctx->time_base.den *= 2;
3991
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3992
                      s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
3993
        }
3994
    }
3995

    
3996
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3997

    
3998
    h->mb_mbaff = 0;
3999
    h->mb_aff_frame = 0;
4000
    last_pic_structure = s0->picture_structure;
4001
    if(h->sps.frame_mbs_only_flag){
4002
        s->picture_structure= PICT_FRAME;
4003
    }else{
4004
        if(get_bits1(&s->gb)) { //field_pic_flag
4005
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
4006
        } else {
4007
            s->picture_structure= PICT_FRAME;
4008
            h->mb_aff_frame = h->sps.mb_aff;
4009
        }
4010
    }
4011

    
4012
    if(h0->current_slice == 0){
4013
        /* See if we have a decoded first field looking for a pair... */
4014
        if (s0->first_field) {
4015
            assert(s0->current_picture_ptr);
4016
            assert(s0->current_picture_ptr->data[0]);
4017
            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
4018

    
4019
            /* figure out if we have a complementary field pair */
4020
            if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
4021
                /*
4022
                 * Previous field is unmatched. Don't display it, but let it
4023
                 * remain for reference if marked as such.
4024
                 */
4025
                s0->current_picture_ptr = NULL;
4026
                s0->first_field = FIELD_PICTURE;
4027

    
4028
            } else {
4029
                if (h->nal_ref_idc &&
4030
                        s0->current_picture_ptr->reference &&
4031
                        s0->current_picture_ptr->frame_num != h->frame_num) {
4032
                    /*
4033
                     * This and previous field were reference, but had
4034
                     * different frame_nums. Consider this field first in
4035
                     * pair. Throw away previous field except for reference
4036
                     * purposes.
4037
                     */
4038
                    s0->first_field = 1;
4039
                    s0->current_picture_ptr = NULL;
4040

    
4041
                } else {
4042
                    /* Second field in complementary pair */
4043
                    s0->first_field = 0;
4044
                }
4045
            }
4046

    
4047
        } else {
4048
            /* Frame or first field in a potentially complementary pair */
4049
            assert(!s0->current_picture_ptr);
4050
            s0->first_field = FIELD_PICTURE;
4051
        }
4052

    
4053
        if((!FIELD_PICTURE || s0->first_field) && frame_start(h) < 0) {
4054
            s0->first_field = 0;
4055
            return -1;
4056
        }
4057
    }
4058
    if(h != h0)
4059
        clone_slice(h, h0);
4060

    
4061
    s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
4062

    
4063
    assert(s->mb_num == s->mb_width * s->mb_height);
4064
    if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
4065
       first_mb_in_slice                    >= s->mb_num){
4066
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
4067
        return -1;
4068
    }
4069
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
4070
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
4071
    if (s->picture_structure == PICT_BOTTOM_FIELD)
4072
        s->resync_mb_y = s->mb_y = s->mb_y + 1;
4073
    assert(s->mb_y < s->mb_height);
4074

    
4075
    if(s->picture_structure==PICT_FRAME){
4076
        h->curr_pic_num=   h->frame_num;
4077
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
4078
    }else{
4079
        h->curr_pic_num= 2*h->frame_num + 1;
4080
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
4081
    }
4082

    
4083
    if(h->nal_unit_type == NAL_IDR_SLICE){
4084
        get_ue_golomb(&s->gb); /* idr_pic_id */
4085
    }
4086

    
4087
    if(h->sps.poc_type==0){
4088
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
4089

    
4090
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
4091
            h->delta_poc_bottom= get_se_golomb(&s->gb);
4092
        }
4093
    }
4094

    
4095
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
4096
        h->delta_poc[0]= get_se_golomb(&s->gb);
4097

    
4098
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
4099
            h->delta_poc[1]= get_se_golomb(&s->gb);
4100
    }
4101

    
4102
    init_poc(h);
4103

    
4104
    if(h->pps.redundant_pic_cnt_present){
4105
        h->redundant_pic_count= get_ue_golomb(&s->gb);
4106
    }
4107

    
4108
    //set defaults, might be overriden a few line later
4109
    h->ref_count[0]= h->pps.ref_count[0];
4110
    h->ref_count[1]= h->pps.ref_count[1];
4111

    
4112
    if(h->slice_type == FF_P_TYPE || h->slice_type == FF_SP_TYPE || h->slice_type == FF_B_TYPE){
4113
        if(h->slice_type == FF_B_TYPE){
4114
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
4115
            if(FIELD_PICTURE && h->direct_spatial_mv_pred)
4116
                av_log(h->s.avctx, AV_LOG_ERROR, "PAFF + spatial direct mode is not implemented\n");
4117
        }
4118
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
4119

    
4120
        if(num_ref_idx_active_override_flag){
4121
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
4122
            if(h->slice_type==FF_B_TYPE)
4123
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
4124

    
4125
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
4126
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
4127
                h->ref_count[0]= h->ref_count[1]= 1;
4128
                return -1;
4129
            }
4130
        }
4131
        if(h->slice_type == FF_B_TYPE)
4132
            h->list_count= 2;
4133
        else
4134
            h->list_count= 1;
4135
    }else
4136
        h->list_count= 0;
4137

    
4138
    if(!default_ref_list_done){
4139
        fill_default_ref_list(h);
4140
    }
4141

    
4142
    if(decode_ref_pic_list_reordering(h) < 0)
4143
        return -1;
4144

    
4145
    if(   (h->pps.weighted_pred          && (h->slice_type == FF_P_TYPE || h->slice_type == FF_SP_TYPE ))
4146
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==FF_B_TYPE ) )
4147
        pred_weight_table(h);
4148
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type==FF_B_TYPE)
4149
        implicit_weight_table(h);
4150
    else
4151
        h->use_weight = 0;
4152

    
4153
    if(h->nal_ref_idc)
4154
        decode_ref_pic_marking(h0, &s->gb);
4155

    
4156
    if(FRAME_MBAFF)
4157
        fill_mbaff_ref_list(h);
4158

    
4159
    if( h->slice_type != FF_I_TYPE && h->slice_type != FF_SI_TYPE && h->pps.cabac ){
4160
        tmp = get_ue_golomb(&s->gb);
4161
        if(tmp > 2){
4162
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
4163
            return -1;
4164
        }
4165
        h->cabac_init_idc= tmp;
4166
    }
4167

    
4168
    h->last_qscale_diff = 0;
4169
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
4170
    if(tmp>51){
4171
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
4172
        return -1;
4173
    }
4174
    s->qscale= tmp;
4175
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
4176
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
4177
    //FIXME qscale / qp ... stuff
4178
    if(h->slice_type == FF_SP_TYPE){
4179
        get_bits1(&s->gb); /* sp_for_switch_flag */
4180
    }
4181
    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
4182
        get_se_golomb(&s->gb); /* slice_qs_delta */
4183
    }
4184

    
4185
    h->deblocking_filter = 1;
4186
    h->slice_alpha_c0_offset = 0;
4187
    h->slice_beta_offset = 0;
4188
    if( h->pps.deblocking_filter_parameters_present ) {
4189
        tmp= get_ue_golomb(&s->gb);
4190
        if(tmp > 2){
4191
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
4192
            return -1;
4193
        }
4194
        h->deblocking_filter= tmp;
4195
        if(h->deblocking_filter < 2)
4196
            h->deblocking_filter^= 1; // 1<->0
4197

    
4198
        if( h->deblocking_filter ) {
4199
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
4200
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
4201
        }
4202
    }
4203

    
4204
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
4205
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type != FF_I_TYPE)
4206
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type == FF_B_TYPE)
4207
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
4208
        h->deblocking_filter= 0;
4209

    
4210
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
4211
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
4212
            /* Cheat slightly for speed:
4213
               Do not bother to deblock across slices. */
4214
            h->deblocking_filter = 2;
4215
        } else {
4216
            h0->max_contexts = 1;
4217
            if(!h0->single_decode_warning) {
4218
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
4219
                h0->single_decode_warning = 1;
4220
            }
4221
            if(h != h0)
4222
                return 1; // deblocking switched inside frame
4223
        }
4224
    }
4225

    
4226
#if 0 //FMO
4227
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
4228
        slice_group_change_cycle= get_bits(&s->gb, ?);
4229
#endif
4230

    
4231
    h0->last_slice_type = slice_type;
4232
    h->slice_num = ++h0->current_slice;
4233

    
4234
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
4235
    h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
4236

    
4237
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4238
        av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
4239
               h->slice_num,
4240
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
4241
               first_mb_in_slice,
4242
               av_get_pict_type_char(h->slice_type),
4243
               pps_id, h->frame_num,
4244
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4245
               h->ref_count[0], h->ref_count[1],
4246
               s->qscale,
4247
               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4248
               h->use_weight,
4249
               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
4250
               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
4251
               );
4252
    }
4253

    
4254
    return 0;
4255
}
4256

    
4257
/**
4258
 *
4259
 */
4260
static inline int get_level_prefix(GetBitContext *gb){
4261
    unsigned int buf;
4262
    int log;
4263

    
4264
    OPEN_READER(re, gb);
4265
    UPDATE_CACHE(re, gb);
4266
    buf=GET_CACHE(re, gb);
4267

    
4268
    log= 32 - av_log2(buf);
4269
#ifdef TRACE
4270
    print_bin(buf>>(32-log), log);
4271
    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__);
4272
#endif
4273

    
4274
    LAST_SKIP_BITS(re, gb, log);
4275
    CLOSE_READER(re, gb);
4276

    
4277
    return log-1;
4278
}
4279

    
4280
static inline int get_dct8x8_allowed(H264Context *h){
4281
    int i;
4282
    for(i=0; i<4; i++){
4283
        if(!IS_SUB_8X8(h->sub_mb_type[i])
4284
           || (!h->sps.direct_8x8_inference_flag && IS_DIRECT(h->sub_mb_type[i])))
4285
            return 0;
4286
    }
4287
    return 1;
4288
}
4289

    
4290
/**
4291
 * decodes a residual block.
4292
 * @param n block index
4293
 * @param scantable scantable
4294
 * @param max_coeff number of coefficients in the block
4295
 * @return <0 if an error occurred
4296
 */
4297
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
4298
    MpegEncContext * const s = &h->s;
4299
    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};
4300
    int level[16];
4301
    int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;
4302

    
4303
    //FIXME put trailing_onex into the context
4304

    
4305
    if(n == CHROMA_DC_BLOCK_INDEX){
4306
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4307
        total_coeff= coeff_token>>2;
4308
    }else{
4309
        if(n == LUMA_DC_BLOCK_INDEX){
4310
            total_coeff= pred_non_zero_count(h, 0);
4311
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4312
            total_coeff= coeff_token>>2;
4313
        }else{
4314
            total_coeff= pred_non_zero_count(h, n);
4315
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4316
            total_coeff= coeff_token>>2;
4317
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4318
        }
4319
    }
4320

    
4321
    //FIXME set last_non_zero?
4322

    
4323
    if(total_coeff==0)
4324
        return 0;
4325
    if(total_coeff > (unsigned)max_coeff) {
4326
        av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
4327
        return -1;
4328
    }
4329

    
4330
    trailing_ones= coeff_token&3;
4331
    tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
4332
    assert(total_coeff<=16);
4333

    
4334
    for(i=0; i<trailing_ones; i++){
4335
        level[i]= 1 - 2*get_bits1(gb);
4336
    }
4337

    
4338
    if(i<total_coeff) {
4339
        int level_code, mask;
4340
        int suffix_length = total_coeff > 10 && trailing_ones < 3;
4341
        int prefix= get_level_prefix(gb);
4342

    
4343
        //first coefficient has suffix_length equal to 0 or 1
4344
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4345
            if(suffix_length)
4346
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4347
            else
4348
                level_code= (prefix<<suffix_length); //part
4349
        }else if(prefix==14){
4350
            if(suffix_length)
4351
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4352
            else
4353
                level_code= prefix + get_bits(gb, 4); //part
4354
        }else{
4355
            level_code= (15<<suffix_length) + get_bits(gb, prefix-3); //part
4356
            if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
4357
            if(prefix>=16)
4358
                level_code += (1<<(prefix-3))-4096;
4359
        }
4360

    
4361
        if(trailing_ones < 3) level_code += 2;
4362

    
4363
        suffix_length = 1;
4364
        if(level_code > 5)
4365
            suffix_length++;
4366
        mask= -(level_code&1);
4367
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
4368
        i++;
4369

    
4370
        //remaining coefficients have suffix_length > 0
4371
        for(;i<total_coeff;i++) {
4372
            static const int suffix_limit[7] = {0,5,11,23,47,95,INT_MAX };
4373
            prefix = get_level_prefix(gb);
4374
            if(prefix<15){
4375
                level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
4376
            }else{
4377
                level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
4378
                if(prefix>=16)
4379
                    level_code += (1<<(prefix-3))-4096;
4380
            }
4381
            mask= -(level_code&1);
4382
            level[i]= (((2+level_code)>>1) ^ mask) - mask;
4383
            if(level_code > suffix_limit[suffix_length])
4384
                suffix_length++;
4385
        }
4386
    }
4387

    
4388
    if(total_coeff == max_coeff)
4389
        zeros_left=0;
4390
    else{
4391
        if(n == CHROMA_DC_BLOCK_INDEX)
4392
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4393
        else
4394
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4395
    }
4396

    
4397
    coeff_num = zeros_left + total_coeff - 1;
4398
    j = scantable[coeff_num];
4399
    if(n > 24){
4400
        block[j] = level[0];
4401
        for(i=1;i<total_coeff;i++) {
4402
            if(zeros_left <= 0)
4403
                run_before = 0;
4404
            else if(zeros_left < 7){
4405
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4406
            }else{
4407
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4408
            }
4409
            zeros_left -= run_before;
4410
            coeff_num -= 1 + run_before;
4411
            j= scantable[ coeff_num ];
4412

    
4413
            block[j]= level[i];
4414
        }
4415
    }else{
4416
        block[j] = (level[0] * qmul[j] + 32)>>6;
4417
        for(i=1;i<total_coeff;i++) {
4418
            if(zeros_left <= 0)
4419
                run_before = 0;
4420
            else if(zeros_left < 7){
4421
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4422
            }else{
4423
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4424
            }
4425
            zeros_left -= run_before;
4426
            coeff_num -= 1 + run_before;
4427
            j= scantable[ coeff_num ];
4428

    
4429
            block[j]= (level[i] * qmul[j] + 32)>>6;
4430
        }
4431
    }
4432

    
4433
    if(zeros_left<0){
4434
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4435
        return -1;
4436
    }
4437

    
4438
    return 0;
4439
}
4440

    
4441
static void predict_field_decoding_flag(H264Context *h){
4442
    MpegEncContext * const s = &h->s;
4443
    const int mb_xy= h->mb_xy;
4444
    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
4445
                ? s->current_picture.mb_type[mb_xy-1]
4446
                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
4447
                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
4448
                : 0;
4449
    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4450
}
4451

    
4452
/**
4453
 * decodes a P_SKIP or B_SKIP macroblock
4454
 */
4455
static void decode_mb_skip(H264Context *h){
4456
    MpegEncContext * const s = &h->s;
4457
    const int mb