Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264_direct.c @ 923bcb40

History | View | Annotate | Download (24.7 KB)

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

    
22
/**
23
 * @file libavcodec/h264_direct.c
24
 * H.264 / AVC / MPEG4 part10 direct mb/block decoding.
25
 * @author Michael Niedermayer <michaelni@gmx.at>
26
 */
27

    
28
#include "internal.h"
29
#include "dsputil.h"
30
#include "avcodec.h"
31
#include "mpegvideo.h"
32
#include "h264.h"
33
#include "rectangle.h"
34

    
35
//#undef NDEBUG
36
#include <assert.h>
37

    
38

    
39
static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
40
    int poc0 = h->ref_list[0][i].poc;
41
    int td = av_clip(poc1 - poc0, -128, 127);
42
    if(td == 0 || h->ref_list[0][i].long_ref){
43
        return 256;
44
    }else{
45
        int tb = av_clip(poc - poc0, -128, 127);
46
        int tx = (16384 + (FFABS(td) >> 1)) / td;
47
        return av_clip((tb*tx + 32) >> 6, -1024, 1023);
48
    }
49
}
50

    
51
void ff_h264_direct_dist_scale_factor(H264Context * const h){
52
    MpegEncContext * const s = &h->s;
53
    const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
54
    const int poc1 = h->ref_list[1][0].poc;
55
    int i, field;
56
    for(field=0; field<2; field++){
57
        const int poc  = h->s.current_picture_ptr->field_poc[field];
58
        const int poc1 = h->ref_list[1][0].field_poc[field];
59
        for(i=0; i < 2*h->ref_count[0]; i++)
60
            h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
61
    }
62

    
63
    for(i=0; i<h->ref_count[0]; i++){
64
        h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
65
    }
66
}
67

    
68
static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
69
    MpegEncContext * const s = &h->s;
70
    Picture * const ref1 = &h->ref_list[1][0];
71
    int j, old_ref, rfield;
72
    int start= mbafi ? 16                      : 0;
73
    int end  = mbafi ? 16+2*h->ref_count[0]    : h->ref_count[0];
74
    int interl= mbafi || s->picture_structure != PICT_FRAME;
75

    
76
    /* bogus; fills in for missing frames */
77
    memset(map[list], 0, sizeof(map[list]));
78

    
79
    for(rfield=0; rfield<2; rfield++){
80
        for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
81
            int poc = ref1->ref_poc[colfield][list][old_ref];
82

    
83
            if     (!interl)
84
                poc |= 3;
85
            else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
86
                poc= (poc&~3) + rfield + 1;
87

    
88
            for(j=start; j<end; j++){
89
                if(4*h->ref_list[0][j].frame_num + (h->ref_list[0][j].reference&3) == poc){
90
                    int cur_ref= mbafi ? (j-16)^field : j;
91
                    map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
92
                    if(rfield == field || !interl)
93
                        map[list][old_ref] = cur_ref;
94
                    break;
95
                }
96
            }
97
        }
98
    }
99
}
100

    
101
void ff_h264_direct_ref_list_init(H264Context * const h){
102
    MpegEncContext * const s = &h->s;
103
    Picture * const ref1 = &h->ref_list[1][0];
104
    Picture * const cur = s->current_picture_ptr;
105
    int list, j, field;
106
    int sidx= (s->picture_structure&1)^1;
107
    int ref1sidx= (ref1->reference&1)^1;
108

    
109
    for(list=0; list<2; list++){
110
        cur->ref_count[sidx][list] = h->ref_count[list];
111
        for(j=0; j<h->ref_count[list]; j++)
112
            cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
113
    }
114

    
115
    if(s->picture_structure == PICT_FRAME){
116
        memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
117
        memcpy(cur->ref_poc  [1], cur->ref_poc  [0], sizeof(cur->ref_poc  [0]));
118
    }
119

    
120
    cur->mbaff= FRAME_MBAFF;
121

    
122
    h->col_fieldoff= 0;
123
    if(s->picture_structure == PICT_FRAME){
124
        int cur_poc = s->current_picture_ptr->poc;
125
        int *col_poc = h->ref_list[1]->field_poc;
126
        h->col_parity= (FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc));
127
        ref1sidx=sidx= h->col_parity;
128
    }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){ // FL -> FL & differ parity
129
        h->col_fieldoff= s->mb_stride*(2*(h->ref_list[1][0].reference) - 3);
130
    }
131

    
132
    if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
133
        return;
134

    
135
    for(list=0; list<2; list++){
136
        fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
137
        if(FRAME_MBAFF)
138
        for(field=0; field<2; field++)
139
            fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
140
    }
141
}
142

    
143
static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
144
    MpegEncContext * const s = &h->s;
145
    int b8_stride = h->b8_stride;
146
    int b4_stride = h->b_stride;
147
    int mb_xy = h->mb_xy;
148
    int mb_type_col[2];
149
    const int16_t (*l1mv0)[2], (*l1mv1)[2];
150
    const int8_t *l1ref0, *l1ref1;
151
    const int is_b8x8 = IS_8X8(*mb_type);
152
    unsigned int sub_mb_type;
153
    int i8, i4;
154
    int ref[2];
155
    int mv[2];
156
    int list;
157

    
158
    assert(h->ref_list[1][0].reference&3);
159

    
160
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
161

    
162
    *mb_type |= MB_TYPE_L0L1;
163
    sub_mb_type |= MB_TYPE_L0L1;
164

    
165
    /* ref = min(neighbors) */
166
    for(list=0; list<2; list++){
167
        int left_ref = h->ref_cache[list][scan8[0] - 1];
168
        int top_ref  = h->ref_cache[list][scan8[0] - 8];
169
        int refc = h->ref_cache[list][scan8[0] - 8 + 4];
170
        const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4];
171
        if(refc == PART_NOT_AVAILABLE){
172
            refc = h->ref_cache[list][scan8[0] - 8 - 1];
173
            C    = h-> mv_cache[list][scan8[0] - 8 - 1];
174
        }
175
        ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc);
176
        if(ref[list] >= 0){
177
            //this is just pred_motion() but with the cases removed that cannot happen for direct blocks
178
            const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
179
            const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
180

    
181
            int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
182
            if(match_count > 1){ //most common
183
                mv[list]= (mid_pred(A[0], B[0], C[0])&0xFFFF)
184
                            +(mid_pred(A[1], B[1], C[1])<<16);
185
            }else {
186
                assert(match_count==1);
187
                if(left_ref==ref[list]){
188
                    mv[list]= *(uint32_t*)A;
189
                }else if(top_ref==ref[list]){
190
                    mv[list]= *(uint32_t*)B;
191
                }else{
192
                    mv[list]= *(uint32_t*)C;
193
                }
194
            }
195
        }else{
196
            int mask= ~(MB_TYPE_L0 << (2*list));
197
            mv[list] = 0;
198
            ref[list] = -1;
199
            if(!is_b8x8)
200
                *mb_type &= mask;
201
            sub_mb_type &= mask;
202
        }
203
    }
204
    if(ref[0] < 0 && ref[1] < 0){
205
        ref[0] = ref[1] = 0;
206
        if(!is_b8x8)
207
            *mb_type |= MB_TYPE_L0L1;
208
        sub_mb_type |= MB_TYPE_L0L1;
209
    }
210

    
211
    if(!is_b8x8 && (mv[0]|mv[1]) == 0){
212
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
213
        fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
214
        fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
215
        fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
216
        *mb_type= (*mb_type & ~(MB_TYPE_8x8|MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_P1L0|MB_TYPE_P1L1))|MB_TYPE_16x16|MB_TYPE_DIRECT2;
217
        return;
218
    }
219

    
220
    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
221
        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
222
            mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
223
            b8_stride = 0;
224
        }else{
225
            mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity
226
        }
227
        goto single_col;
228
    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
229
        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
230
            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
231
            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
232
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
233
            b8_stride *= 3;
234
            b4_stride *= 6;
235

    
236
            sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
237
            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
238
                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
239
                && !is_b8x8){
240
                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */
241
            }else{
242
                *mb_type   |= MB_TYPE_8x8;
243
            }
244
        }else{                                           //     AFR/FR    -> AFR/FR
245
single_col:
246
            mb_type_col[0] =
247
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
248

    
249
            sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
250
            if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
251
                *mb_type   |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */
252
            }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
253
                *mb_type   |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
254
            }else{
255
                if(!h->sps.direct_8x8_inference_flag){
256
                    /* FIXME save sub mb types from previous frames (or derive from MVs)
257
                    * so we know exactly what block size to use */
258
                    sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */
259
                }
260
                *mb_type   |= MB_TYPE_8x8;
261
            }
262
        }
263
    }
264

    
265
    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
266
    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
267
    l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
268
    l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
269
    if(!b8_stride){
270
        if(s->mb_y&1){
271
            l1ref0 += h->b8_stride;
272
            l1ref1 += h->b8_stride;
273
            l1mv0  +=  2*b4_stride;
274
            l1mv1  +=  2*b4_stride;
275
        }
276
    }
277

    
278

    
279
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
280
            int n=0;
281
            for(i8=0; i8<4; i8++){
282
                int x8 = i8&1;
283
                int y8 = i8>>1;
284
                int xy8 = x8+y8*b8_stride;
285
                int xy4 = 3*x8+y8*b4_stride;
286
                int a,b;
287

    
288
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
289
                    continue;
290
                h->sub_mb_type[i8] = sub_mb_type;
291

    
292
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
293
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
294
                if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
295
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
296
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
297
                    a=b=0;
298
                    if(ref[0] > 0)
299
                        a= mv[0];
300
                    if(ref[1] > 0)
301
                        b= mv[1];
302
                    n++;
303
                }else{
304
                    a= mv[0];
305
                    b= mv[1];
306
                }
307
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
308
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
309
            }
310
            if(!is_b8x8 && !(n&3))
311
                *mb_type= (*mb_type & ~(MB_TYPE_8x8|MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_P1L0|MB_TYPE_P1L1))|MB_TYPE_16x16|MB_TYPE_DIRECT2;
312
        }else if(IS_16X16(*mb_type)){
313
            int a,b;
314

    
315
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
316
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
317
            if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref
318
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
319
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
320
                       && h->x264_build>33U))){
321
                a=b=0;
322
                if(ref[0] > 0)
323
                    a= mv[0];
324
                if(ref[1] > 0)
325
                    b= mv[1];
326
            }else{
327
                a= mv[0];
328
                b= mv[1];
329
            }
330
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
331
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
332
        }else{
333
            int n=0;
334
            for(i8=0; i8<4; i8++){
335
                const int x8 = i8&1;
336
                const int y8 = i8>>1;
337

    
338
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
339
                    continue;
340
                h->sub_mb_type[i8] = sub_mb_type;
341

    
342
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4);
343
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4);
344
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
345
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
346

    
347
                /* col_zero_flag */
348
                if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && (   l1ref0[x8 + y8*b8_stride] == 0
349
                                              || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
350
                                                  && h->x264_build>33U))){
351
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
352
                    if(IS_SUB_8X8(sub_mb_type)){
353
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
354
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
355
                            if(ref[0] == 0)
356
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
357
                            if(ref[1] == 0)
358
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
359
                            n+=4;
360
                        }
361
                    }else{
362
                        int m=0;
363
                    for(i4=0; i4<4; i4++){
364
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
365
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
366
                            if(ref[0] == 0)
367
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
368
                            if(ref[1] == 0)
369
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
370
                            m++;
371
                        }
372
                    }
373
                    if(!(m&3))
374
                        h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8;
375
                    n+=m;
376
                    }
377
                }
378
            }
379
            if(!is_b8x8 && !(n&15))
380
                *mb_type= (*mb_type & ~(MB_TYPE_8x8|MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_P1L0|MB_TYPE_P1L1))|MB_TYPE_16x16|MB_TYPE_DIRECT2;
381
        }
382
}
383

    
384
static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
385
    MpegEncContext * const s = &h->s;
386
    int b8_stride = h->b8_stride;
387
    int b4_stride = h->b_stride;
388
    int mb_xy = h->mb_xy;
389
    int mb_type_col[2];
390
    const int16_t (*l1mv0)[2], (*l1mv1)[2];
391
    const int8_t *l1ref0, *l1ref1;
392
    const int is_b8x8 = IS_8X8(*mb_type);
393
    unsigned int sub_mb_type;
394
    int i8, i4;
395

    
396
    assert(h->ref_list[1][0].reference&3);
397

    
398
    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
399
        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
400
            mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
401
            b8_stride = 0;
402
        }else{
403
            mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity
404
        }
405
        goto single_col;
406
    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
407
        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
408
            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
409
            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
410
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
411
            b8_stride *= 3;
412
            b4_stride *= 6;
413

    
414
            sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
415

    
416
            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
417
                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
418
                && !is_b8x8){
419
                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
420
            }else{
421
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
422
            }
423
        }else{                                           //     AFR/FR    -> AFR/FR
424
single_col:
425
            mb_type_col[0] =
426
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
427

    
428
            sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
429
            if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
430
                *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
431
            }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
432
                *mb_type   |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
433
            }else{
434
                if(!h->sps.direct_8x8_inference_flag){
435
                    /* FIXME save sub mb types from previous frames (or derive from MVs)
436
                    * so we know exactly what block size to use */
437
                    sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
438
                }
439
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
440
            }
441
        }
442
    }
443

    
444
    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
445
    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
446
    l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
447
    l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
448
    if(!b8_stride){
449
        if(s->mb_y&1){
450
            l1ref0 += h->b8_stride;
451
            l1ref1 += h->b8_stride;
452
            l1mv0  +=  2*b4_stride;
453
            l1mv1  +=  2*b4_stride;
454
        }
455
    }
456

    
457
    {
458
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
459
        const int *dist_scale_factor = h->dist_scale_factor;
460
        int ref_offset;
461

    
462
        if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
463
            map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
464
            map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
465
            dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];
466
        }
467
        ref_offset = (h->ref_list[1][0].mbaff<<4) & (mb_type_col[0]>>3); //if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0])) ref_offset=16 else 0
468

    
469
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
470
            int y_shift  = 2*!IS_INTERLACED(*mb_type);
471
            assert(h->sps.direct_8x8_inference_flag);
472

    
473
            for(i8=0; i8<4; i8++){
474
                const int x8 = i8&1;
475
                const int y8 = i8>>1;
476
                int ref0, scale;
477
                const int16_t (*l1mv)[2]= l1mv0;
478

    
479
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
480
                    continue;
481
                h->sub_mb_type[i8] = sub_mb_type;
482

    
483
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
484
                if(IS_INTRA(mb_type_col[y8])){
485
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
486
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
487
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
488
                    continue;
489
                }
490

    
491
                ref0 = l1ref0[x8 + y8*b8_stride];
492
                if(ref0 >= 0)
493
                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
494
                else{
495
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
496
                    l1mv= l1mv1;
497
                }
498
                scale = dist_scale_factor[ref0];
499
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
500

    
501
                {
502
                    const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
503
                    int my_col = (mv_col[1]<<y_shift)/2;
504
                    int mx = (scale * mv_col[0] + 128) >> 8;
505
                    int my = (scale * my_col + 128) >> 8;
506
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
507
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
508
                }
509
            }
510
            return;
511
        }
512

    
513
        /* one-to-one mv scaling */
514

    
515
        if(IS_16X16(*mb_type)){
516
            int ref, mv0, mv1;
517

    
518
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
519
            if(IS_INTRA(mb_type_col[0])){
520
                ref=mv0=mv1=0;
521
            }else{
522
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
523
                                                : map_col_to_list0[1][l1ref1[0] + ref_offset];
524
                const int scale = dist_scale_factor[ref0];
525
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
526
                int mv_l0[2];
527
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
528
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
529
                ref= ref0;
530
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
531
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
532
            }
533
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
534
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
535
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
536
        }else{
537
            for(i8=0; i8<4; i8++){
538
                const int x8 = i8&1;
539
                const int y8 = i8>>1;
540
                int ref0, scale;
541
                const int16_t (*l1mv)[2]= l1mv0;
542

    
543
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
544
                    continue;
545
                h->sub_mb_type[i8] = sub_mb_type;
546
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
547
                if(IS_INTRA(mb_type_col[0])){
548
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
549
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
550
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
551
                    continue;
552
                }
553

    
554
                ref0 = l1ref0[x8 + y8*b8_stride];
555
                if(ref0 >= 0)
556
                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
557
                else{
558
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
559
                    l1mv= l1mv1;
560
                }
561
                scale = dist_scale_factor[ref0];
562

    
563
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
564
                if(IS_SUB_8X8(sub_mb_type)){
565
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
566
                    int mx = (scale * mv_col[0] + 128) >> 8;
567
                    int my = (scale * mv_col[1] + 128) >> 8;
568
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
569
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
570
                }else
571
                for(i4=0; i4<4; i4++){
572
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
573
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
574
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
575
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
576
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
577
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
578
                }
579
            }
580
        }
581
    }
582
}
583

    
584
void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
585
    if(h->direct_spatial_mv_pred){
586
        pred_spatial_direct_motion(h, mb_type);
587
    }else{
588
        pred_temp_direct_motion(h, mb_type);
589
    }
590
}