summaryrefslogtreecommitdiffstats
path: root/libavcodec/h264_direct.c
blob: d22780d925524960cc82e11eb23b5d22b9a0d09a (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
/*
 * H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding
 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file
 * H.264 / AVC / MPEG4 part10 direct mb/block decoding.
 * @author Michael Niedermayer <michaelni@gmx.at>
 */

#include "internal.h"
#include "dsputil.h"
#include "avcodec.h"
#include "mpegvideo.h"
#include "h264.h"
#include "rectangle.h"

//#undef NDEBUG
#include <assert.h>


static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
    int poc0 = h->ref_list[0][i].poc;
    int td = av_clip(poc1 - poc0, -128, 127);
    if(td == 0 || h->ref_list[0][i].long_ref){
        return 256;
    }else{
        int tb = av_clip(poc - poc0, -128, 127);
        int tx = (16384 + (FFABS(td) >> 1)) / td;
        return av_clip((tb*tx + 32) >> 6, -1024, 1023);
    }
}

void ff_h264_direct_dist_scale_factor(H264Context * const h){
    MpegEncContext * const s = &h->s;
    const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
    const int poc1 = h->ref_list[1][0].poc;
    int i, field;
    for(field=0; field<2; field++){
        const int poc  = h->s.current_picture_ptr->field_poc[field];
        const int poc1 = h->ref_list[1][0].field_poc[field];
        for(i=0; i < 2*h->ref_count[0]; i++)
            h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
    }

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

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

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

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

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

            for(j=start; j<end; j++){
                if(4*h->ref_list[0][j].frame_num + (h->ref_list[0][j].reference&3) == poc){
                    int cur_ref= mbafi ? (j-16)^field : j;
                    map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
                    if(rfield == field || !interl)
                        map[list][old_ref] = cur_ref;
                    break;
                }
            }
        }
    }
}

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

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

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

    cur->mbaff= FRAME_MBAFF;

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

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

    for(list=0; list<2; list++){
        fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
        if(FRAME_MBAFF)
        for(field=0; field<2; field++)
            fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
    }
}

static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
    MpegEncContext * const s = &h->s;
    int b8_stride = 2;
    int b4_stride = h->b_stride;
    int mb_xy = h->mb_xy;
    int mb_type_col[2];
    const int16_t (*l1mv0)[2], (*l1mv1)[2];
    const int8_t *l1ref0, *l1ref1;
    const int is_b8x8 = IS_8X8(*mb_type);
    unsigned int sub_mb_type= MB_TYPE_L0L1;
    int i8, i4;
    int ref[2];
    int mv[2];
    int list;

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

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


    /* ref = min(neighbors) */
    for(list=0; list<2; list++){
        int left_ref = h->ref_cache[list][scan8[0] - 1];
        int top_ref  = h->ref_cache[list][scan8[0] - 8];
        int refc = h->ref_cache[list][scan8[0] - 8 + 4];
        const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4];
        if(refc == PART_NOT_AVAILABLE){
            refc = h->ref_cache[list][scan8[0] - 8 - 1];
            C    = h-> mv_cache[list][scan8[0] - 8 - 1];
        }
        ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc);
        if(ref[list] >= 0){
            //this is just pred_motion() but with the cases removed that cannot happen for direct blocks
            const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
            const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];

            int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
            if(match_count > 1){ //most common
                mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]),
                                     mid_pred(A[1], B[1], C[1]) );
            }else {
                assert(match_count==1);
                if(left_ref==ref[list]){
                    mv[list]= AV_RN32A(A);
                }else if(top_ref==ref[list]){
                    mv[list]= AV_RN32A(B);
                }else{
                    mv[list]= AV_RN32A(C);
                }
            }
        }else{
            int mask= ~(MB_TYPE_L0 << (2*list));
            mv[list] = 0;
            ref[list] = -1;
            if(!is_b8x8)
                *mb_type &= mask;
            sub_mb_type &= mask;
        }
    }
    if(ref[0] < 0 && ref[1] < 0){
        ref[0] = ref[1] = 0;
        if(!is_b8x8)
            *mb_type |= MB_TYPE_L0L1;
        sub_mb_type |= MB_TYPE_L0L1;
    }

    if(!(is_b8x8|mv[0]|mv[1])){
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
        fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
        fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
        fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
        *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;
        return;
    }

    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
            mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
            b8_stride = 0;
        }else{
            mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity
        }
        goto single_col;
    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
            b8_stride = 2+4*s->mb_stride;
            b4_stride *= 6;

            sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
                && !is_b8x8){
                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */
            }else{
                *mb_type   |= MB_TYPE_8x8;
            }
        }else{                                           //     AFR/FR    -> AFR/FR
single_col:
            mb_type_col[0] =
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];

            sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
            if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
                *mb_type   |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */
            }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
                *mb_type   |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
            }else{
                if(!h->sps.direct_8x8_inference_flag){
                    /* FIXME save sub mb types from previous frames (or derive from MVs)
                    * so we know exactly what block size to use */
                    sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */
                }
                *mb_type   |= MB_TYPE_8x8;
            }
        }
    }

    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
    l1ref0 = &h->ref_list[1][0].ref_index [0][4*mb_xy];
    l1ref1 = &h->ref_list[1][0].ref_index [1][4*mb_xy];
    if(!b8_stride){
        if(s->mb_y&1){
            l1ref0 += 2;
            l1ref1 += 2;
            l1mv0  +=  2*b4_stride;
            l1mv1  +=  2*b4_stride;
        }
    }


        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
            int n=0;
            for(i8=0; i8<4; i8++){
                int x8 = i8&1;
                int y8 = i8>>1;
                int xy8 = x8+y8*b8_stride;
                int xy4 = 3*x8+y8*b4_stride;
                int a,b;

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

                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
                if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
                    a=b=0;
                    if(ref[0] > 0)
                        a= mv[0];
                    if(ref[1] > 0)
                        b= mv[1];
                    n++;
                }else{
                    a= mv[0];
                    b= mv[1];
                }
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
            }
            if(!is_b8x8 && !(n&3))
                *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;
        }else if(IS_16X16(*mb_type)){
            int a,b;

            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
            if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
                       && h->x264_build>33U))){
                a=b=0;
                if(ref[0] > 0)
                    a= mv[0];
                if(ref[1] > 0)
                    b= mv[1];
            }else{
                a= mv[0];
                b= mv[1];
            }
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
        }else{
            int n=0;
            for(i8=0; i8<4; i8++){
                const int x8 = i8&1;
                const int y8 = i8>>1;

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

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

                assert(b8_stride==2);
                /* col_zero_flag */
                if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && (   l1ref0[i8] == 0
                                              || (l1ref0[i8] < 0 && l1ref1[i8] == 0
                                                  && h->x264_build>33U))){
                    const int16_t (*l1mv)[2]= l1ref0[i8] == 0 ? l1mv0 : l1mv1;
                    if(IS_SUB_8X8(sub_mb_type)){
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
                            if(ref[0] == 0)
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
                            if(ref[1] == 0)
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
                            n+=4;
                        }
                    }else{
                        int m=0;
                    for(i4=0; i4<4; i4++){
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
                            if(ref[0] == 0)
                                AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]);
                            if(ref[1] == 0)
                                AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]);
                            m++;
                        }
                    }
                    if(!(m&3))
                        h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8;
                    n+=m;
                    }
                }
            }
            if(!is_b8x8 && !(n&15))
                *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;
        }
}

static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
    MpegEncContext * const s = &h->s;
    int b8_stride = 2;
    int b4_stride = h->b_stride;
    int mb_xy = h->mb_xy;
    int mb_type_col[2];
    const int16_t (*l1mv0)[2], (*l1mv1)[2];
    const int8_t *l1ref0, *l1ref1;
    const int is_b8x8 = IS_8X8(*mb_type);
    unsigned int sub_mb_type;
    int i8, i4;

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

    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
            mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
            b8_stride = 0;
        }else{
            mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity
        }
        goto single_col;
    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
            b8_stride = 2+4*s->mb_stride;
            b4_stride *= 6;

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

            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
                && !is_b8x8){
                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
            }else{
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
            }
        }else{                                           //     AFR/FR    -> AFR/FR
single_col:
            mb_type_col[0] =
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];

            sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
            if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
                *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
            }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
                *mb_type   |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
            }else{
                if(!h->sps.direct_8x8_inference_flag){
                    /* FIXME save sub mb types from previous frames (or derive from MVs)
                    * so we know exactly what block size to use */
                    sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
                }
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
            }
        }
    }

    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
    l1ref0 = &h->ref_list[1][0].ref_index [0][4*mb_xy];
    l1ref1 = &h->ref_list[1][0].ref_index [1][4*mb_xy];
    if(!b8_stride){
        if(s->mb_y&1){
            l1ref0 += 2;
            l1ref1 += 2;
            l1mv0  +=  2*b4_stride;
            l1mv1  +=  2*b4_stride;
        }
    }

    {
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
        const int *dist_scale_factor = h->dist_scale_factor;
        int ref_offset;

        if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
            map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
            map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
            dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];
        }
        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

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

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

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

                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
                if(IS_INTRA(mb_type_col[y8])){
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
                    continue;
                }

                ref0 = l1ref0[x8 + y8*b8_stride];
                if(ref0 >= 0)
                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
                else{
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
                    l1mv= l1mv1;
                }
                scale = dist_scale_factor[ref0];
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);

                {
                    const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
                    int my_col = (mv_col[1]<<y_shift)/2;
                    int mx = (scale * mv_col[0] + 128) >> 8;
                    int my = (scale * my_col + 128) >> 8;
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
                }
            }
            return;
        }

        /* one-to-one mv scaling */

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

            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
            if(IS_INTRA(mb_type_col[0])){
                ref=mv0=mv1=0;
            }else{
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
                                                : map_col_to_list0[1][l1ref1[0] + ref_offset];
                const int scale = dist_scale_factor[ref0];
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
                int mv_l0[2];
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
                ref= ref0;
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
            }
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
        }else{
            for(i8=0; i8<4; i8++){
                const int x8 = i8&1;
                const int y8 = i8>>1;
                int ref0, scale;
                const int16_t (*l1mv)[2]= l1mv0;

                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
                    continue;
                h->sub_mb_type[i8] = sub_mb_type;
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
                if(IS_INTRA(mb_type_col[0])){
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
                    continue;
                }

                assert(b8_stride == 2);
                ref0 = l1ref0[i8];
                if(ref0 >= 0)
                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
                else{
                    ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset];
                    l1mv= l1mv1;
                }
                scale = dist_scale_factor[ref0];

                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
                if(IS_SUB_8X8(sub_mb_type)){
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
                    int mx = (scale * mv_col[0] + 128) >> 8;
                    int my = (scale * mv_col[1] + 128) >> 8;
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
                }else
                for(i4=0; i4<4; i4++){
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
                    AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]],
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]));
                }
            }
        }
    }
}

void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
    if(h->direct_spatial_mv_pred){
        pred_spatial_direct_motion(h, mb_type);
    }else{
        pred_temp_direct_motion(h, mb_type);
    }
}
OpenPOWER on IntegriCloud